Merge branch 'master' into add-nifti-support

2018-07-20 13:57:51 +01:00 · 2018-07-20 13:57:51 +01:00 · 3201b4373d
parent d03416d386 d06c2fee22
commit 3201b4373d
56 changed files with 4227 additions and 160 deletions
--- a/.gitignore
+++ b/.gitignore
@ -1,5 +1,6 @@
 .pytest_cache
 compile
+.pytest_cache
 a.out
 *.log
 *.trs
--- a/6
+++ b/6
@ -31,7 +31,13 @@
 - add Mitchell kernel
 - pyramid builders have a choice of 2x2 shrinkers [harukizaemon]
 - add `palette` option to pngsave [felixbuenemann]
+<<<<<<< HEAD
 - add basic nifti load/save support
+=======
+- support writing string-valued fields via libexif
+- paste in the test suite from pyvips
+- get EXIF tag names from tag plus ifd [@Nan619]
+>>>>>>> master

 12/3/18 started 8.6.4
 - better fitting of fonts with overhanging edges [Adrià]
--- a/README.md
+++ b/README.md
@ -8,28 +8,25 @@ similar libraries, [libvips runs quickly and uses little
 memory](https://github.com/jcupitt/libvips/wiki/Speed-and-memory-use).
 libvips is licensed under the LGPL 2.1+.

-It has around 300 operations covering arithmetic, histograms,
-convolutions, morphological operations, frequency filtering, colour,
-resampling, statistics and others. It supports a large range of numeric
-formats, from 8-bit int to 128-bit complex. It supports a good range of
-image formats, including JPEG, TIFF, PNG, WebP, FITS, Matlab, OpenEXR,
-PDF, SVG, HDR, PPM, CSV, GIF, Analyze, DeepZoom, and OpenSlide.  It can
-also load images via ImageMagick or GraphicsMagick.
+It has around 300 operations covering arithmetic, histograms, convolutions,
+morphological operations, frequency filtering, colour, resampling, statistics
+and others. It supports a large range of numeric formats, from 8-bit int
+to 128-bit complex. It supports a good range of image formats, including
+JPEG, TIFF, PNG, WebP, FITS, Matlab, OpenEXR, PDF, SVG, HDR, PPM, CSV, GIF,
+Analyze, DeepZoom, and OpenSlide.  It can also load images via ImageMagick
+or GraphicsMagick.

 It has APIs for
-[C](http://jcupitt.github.io/libvips/API/current/using-from-c.html)
-and
+[C](http://jcupitt.github.io/libvips/API/current/using-from-c.html) and
 [C++](http://jcupitt.github.io/libvips/API/current/using-from-cpp.html)
 and a [command-line
 interface](http://jcupitt.github.io/libvips/API/current/using-cli.html).
-Bindings are available for 
-[Python](https://pypi.python.org/pypi/pyvips),
+Bindings are available for [Python](https://pypi.python.org/pypi/pyvips),
 [Ruby](https://rubygems.org/gems/ruby-vips),
 [PHP](https://github.com/jcupitt/php-vips),
 [Go](https://github.com/davidbyttow/govips),
-[Lua](https://github.com/jcupitt/lua-vips),
-JavaScript and others. There is full
-[documentation](http://jcupitt.github.io/libvips/API/current).
+[Lua](https://github.com/jcupitt/lua-vips), JavaScript and others. There is
+full [documentation](http://jcupitt.github.io/libvips/API/current).
 There are several GUIs as well, see the [VIPS
 website](http://jcupitt.github.io/libvips).

@ -44,105 +41,120 @@ https://github.com/jcupitt/libvips/releases

 Untar, then in the libvips directory you should just be able to do:

-	$ ./configure
+    $ ./configure

-Check the summary at the end of `configure` carefully. 
-libvips must have `build-essential`, `pkg-config`, `glib2.0-dev`,
-`libexpat1-dev`.
+Check the summary at the end of `configure` carefully.  libvips must have
+`build-essential`, `pkg-config`, `glib2.0-dev`, `libexpat1-dev`.

-You'll need the dev packages for the file format support you
-want. For basic jpeg and tiff support, you'll need `libtiff5-dev`,
-`libjpeg-turbo8-dev`, and `libgsf-1-dev`.  See the **Dependencies** section
-below for a full list of the things that libvips can be configured to use.
+You'll need the dev packages for the file format support you want. For basic
+jpeg and tiff support, you'll need `libtiff5-dev`, `libjpeg-turbo8-dev`,
+and `libgsf-1-dev`.  See the **Dependencies** section below for a full list
+of the things that libvips can be configured to use.

 Once `configure` is looking OK, compile and install with the usual:

-	$ make
-	$ sudo make install
+    $ make
+    $ sudo make install

 By default this will install files to `/usr/local`.

 We have detailed guides on the wiki for [building on
-Windows](https://github.com/jcupitt/libvips/wiki/Build-for-Windows)
-and [building on OS
-X](https://github.com/jcupitt/libvips/wiki/Build-for-macOS).
+Windows](https://github.com/jcupitt/libvips/wiki/Build-for-Windows) and
+[building on OS X](https://github.com/jcupitt/libvips/wiki/Build-for-macOS).
+
+# Testing
+
+Do a basic test of your build with:
+
+    $ make check
+
+Run the libvips test suite with:
+
+    $ pytest
+
+Run a specific test with:
+
+    $ pytest test/test-suite/test_foreign.py -k test_tiff
+
+YOu will need to install a variety of Python packages for this, including
+pyvips, the libvips Python binding.

 # Building libvips from git

 Checkout the latest sources with:

-	$ git clone git://github.com/jcupitt/libvips.git
+    $ git clone git://github.com/jcupitt/libvips.git

 Building from git needs more packages, you'll need at least `swig`, `gtk-doc` 
 and `gobject-introspection`, see the dependencies section below. For example:

-	$ brew install gtk-doc swig
+    $ brew install gtk-doc swig

 Then build the build system with:

-	$ ./autogen.sh
+    $ ./autogen.sh

 Debug build:

-	$ CFLAGS="-g -Wall" CXXFLAGS="-g -Wall" \
-		./configure --prefix=/home/john/vips --enable-debug
-	$ make
-	$ make install
+    $ CFLAGS="-g -Wall" CXXFLAGS="-g -Wall" \
+        ./configure --prefix=/home/john/vips --enable-debug
+    $ make
+    $ make install

 Leak check:

-	$ export G_DEBUG=gc-friendly
-	$ valgrind --suppressions=libvips.supp \
-		--leak-check=yes \
-		vips ... > vips-vg.log 2>&1
+    $ export G_DEBUG=gc-friendly
+    $ valgrind --suppressions=libvips.supp \
+	       --leak-check=yes \
+        vips ... > vips-vg.log 2>&1

 Memory error debug:

-	$ valgrind --vgdb=yes --vgdb-error=0 vips  ...
+    $ valgrind --vgdb=yes --vgdb-error=0 vips  ...

 valgrind threading check:

-	$ valgrind --tool=helgrind vips ... > vips-vg.log 2>&1
+    $ valgrind --tool=helgrind vips ... > vips-vg.log 2>&1

 Clang build:

-	$ CC=clang CXX=clang++ ./configure --prefix=/home/john/vips
+    $ CC=clang CXX=clang++ ./configure --prefix=/home/john/vips

 Clang static analysis:

-	$ scan-build ./configure --disable-introspection --disable-debug
-	$ scan-build -o scan -v make 
-	$ scan-view scan/2013-11-22-2
+    $ scan-build ./configure --disable-introspection --disable-debug
+    $ scan-build -o scan -v make 
+    $ scan-view scan/2013-11-22-2

 Clang dynamic analysis:

-	$ FLAGS="-O1 -g -fsanitize=address"
-	$ FLAGS="$FLAGS -fno-omit-frame-pointer -fno-optimize-sibling-calls"
-	$ CC=clang CXX=clang++ LD=clang \
-		CFLAGS="$FLAGS" CXXFLAGS="$FLAGS" LDFLAGS=-fsanitize=address \
-		./configure --prefix=/home/john/vips 
+    $ FLAGS="-O1 -g -fsanitize=address"
+    $ FLAGS="$FLAGS -fno-omit-frame-pointer -fno-optimize-sibling-calls"
+    $ CC=clang CXX=clang++ LD=clang \
+        CFLAGS="$FLAGS" CXXFLAGS="$FLAGS" LDFLAGS=-fsanitize=address \
+        ./configure --prefix=/home/john/vips 

-	$ FLAGS="-O1 -g -fsanitize=thread"
-	$ FLAGS="$FLAGS -fPIC"
-	$ FLAGS="$FLAGS -fno-omit-frame-pointer -fno-optimize-sibling-calls"
-	$ CC=clang CXX=clang++ LD=clang \
-		CFLAGS="$FLAGS" CXXFLAGS="$FLAGS" \
-		LDFLAGS="-fsanitize=thread -fPIC" \
-		./configure --prefix=/home/john/vips \
-			--without-magick \
-			--disable-introspection
-	$ G_DEBUG=gc-friendly vips copy ~/pics/k2.jpg x.jpg >& log
+    $ FLAGS="-O1 -g -fsanitize=thread"
+    $ FLAGS="$FLAGS -fPIC"
+    $ FLAGS="$FLAGS -fno-omit-frame-pointer -fno-optimize-sibling-calls"
+    $ CC=clang CXX=clang++ LD=clang \
+      CFLAGS="$FLAGS" CXXFLAGS="$FLAGS" \
+      LDFLAGS="-fsanitize=thread -fPIC" \
+      ./configure --prefix=/home/john/vips \
+        --without-magick \
+        --disable-introspection
+    $ G_DEBUG=gc-friendly vips copy ~/pics/k2.jpg x.jpg >& log

 Build with the GCC auto-vectorizer and diagnostics (or just -O3):

-	$ FLAGS="-O2 -march=native -ffast-math"
-	$ FLAGS="$FLAGS -ftree-vectorize -fdump-tree-vect-details"
-	$ CFLAGS="$FLAGS" CXXFLAGS="$FLAGS" \
-		./configure --prefix=/home/john/vips 
+    $ FLAGS="-O2 -march=native -ffast-math"
+    $ FLAGS="$FLAGS -ftree-vectorize -fdump-tree-vect-details"
+    $ CFLAGS="$FLAGS" CXXFLAGS="$FLAGS" \
+      ./configure --prefix=/home/john/vips 

 Static analysis with:

-	$ cppcheck --force --enable=style . &> cppcheck.log
+    $ cppcheck --force --enable=style . &> cppcheck.log

 # Dependencies 

@ -160,19 +172,19 @@ them in the default path and in `$prefix`. If you have installed your own
 versions of these libraries in a different location, libvips will not see
 them. Use switches to libvips configure like:

-	./configure --prefix=/Users/john/vips \
-		--with-giflib-includes=/opt/local/include \
-		--with-giflib-libraries=/opt/local/lib \
-		--with-tiff-includes=/opt/local/include \
-		--with-tiff-libraries=/opt/local/lib \
-		--with-jpeg-includes=/opt/local/include \
-		--with-jpeg-libraries=/opt/local/lib
+    ./configure --prefix=/Users/john/vips \
+        --with-giflib-includes=/opt/local/include \
+        --with-giflib-libraries=/opt/local/lib \
+        --with-tiff-includes=/opt/local/include \
+        --with-tiff-libraries=/opt/local/lib \
+        --with-jpeg-includes=/opt/local/include \
+        --with-jpeg-libraries=/opt/local/lib

 or perhaps:

-	CFLAGS="-g -Wall -I/opt/local/include -L/opt/local/lib" \
-		CXXFLAGS="-g -Wall -I/opt/local/include -L/opt/local/lib" \
-		./configure --without-python --prefix=/Users/john/vips 
+    CFLAGS="-g -Wall -I/opt/local/include -L/opt/local/lib" \
+    CXXFLAGS="-g -Wall -I/opt/local/include -L/opt/local/lib" \
+    ./configure --without-python --prefix=/Users/john/vips 

 to get libvips to see your builds.

--- a/libvips/foreign/exif.c
+++ b/libvips/foreign/exif.c
@ -7,6 +7,10 @@
 * 	- only read orientation from ifd0
 * 1/2/18
 * 	- remove exif thumbnail if "jpeg-thumbnail-data" has been removed
+ * 3/7/18
+ * 	- add support for writing string-valued fields
+ * 9/7/18 [@Nan619]
+ * 	- get tag name from tag plus ifd 
 */

 /*
@ -307,6 +311,21 @@ typedef struct _VipsExifParams {
 	ExifData *ed;
 } VipsExifParams;

+/* tags do not uniquely set tag names: the same tag can have different
+ * names in different ifds.
+ *
+ * As long as this entry has been linked to an ifd, get the tag name.
+ */
+static const char *
+vips_exif_entry_get_name( ExifEntry *entry )
+{
+	if( !entry->parent )
+		return( NULL );
+
+	return( exif_tag_get_name_in_ifd( entry->tag, 
+		exif_entry_get_ifd( entry ) ) );
+}
+
 static void
 vips_exif_attach_entry( ExifEntry *entry, VipsExifParams *params )
 {
@ -316,7 +335,7 @@ vips_exif_attach_entry( ExifEntry *entry, VipsExifParams *params )
 	char value_txt[256];
 	VipsBuf value = VIPS_BUF_STATIC( value_txt );

-	if( !(tag_name = exif_tag_get_name( entry->tag )) )
+	if( !(tag_name = vips_exif_entry_get_name( entry )) )
 		return;

 	vips_buf_appendf( &vips_name, "exif-ifd%d-%s", 
@ -538,7 +557,7 @@ vips_exif_set_int( ExifData *ed,
 	offset = component * sizeof_component;

 	VIPS_DEBUG_MSG( "vips_exif_set_int: %s = %d\n",
-		exif_tag_get_name( entry->tag ), value );
+		vips_exif_entry_get_name( entry ), value );

 	if( entry->format == EXIF_FORMAT_SHORT ) 
 		exif_set_short( entry->data + offset, bo, value );
@ -616,7 +635,7 @@ vips_exif_set_rational( ExifData *ed,
 	offset = component * sizeof_component;

 	VIPS_DEBUG_MSG( "vips_exif_set_rational: %s = \"%s\"\n",
-		exif_tag_get_name( entry->tag ), value );
+		vips_exif_entry_get_name( entry ), value );

 	if( entry->format == EXIF_FORMAT_RATIONAL ) {
 		ExifRational rv;
@ -670,7 +689,7 @@ vips_exif_set_double( ExifData *ed,
 	offset = component * sizeof_component;

 	VIPS_DEBUG_MSG( "vips_exif_set_double: %s = %g\n",
-		exif_tag_get_name( entry->tag ), value );
+		vips_exif_entry_get_name( entry ), value );

 	if( entry->format == EXIF_FORMAT_RATIONAL ) {
 		ExifRational rv;
@ -706,6 +725,183 @@ vips_exif_set_double( ExifData *ed,
 typedef void (*write_fn)( ExifData *ed, 
 	ExifEntry *entry, unsigned long component, void *data );

+/* String-valued tags need special treatment, sadly.
+ *
+ * Strings are written in three ways: 
+ *
+ * 1. As ASCII, but with an 8-byte preamble giving the encoding (it's always
+ * ASCII though) and the format undefined.
+ * 2. As plain ASCII, with the format giving the encoding.
+ * 3. As UTF16 in the MS tags.
+ */
+
+static gboolean
+tag_is_encoding( ExifTag tag )
+{
+	return( tag == EXIF_TAG_USER_COMMENT );
+}
+
+static gboolean
+tag_is_ascii( ExifTag tag )
+{
+	return( tag == EXIF_TAG_MAKE ||
+		tag == EXIF_TAG_MODEL ||
+		tag == EXIF_TAG_IMAGE_DESCRIPTION ||
+		tag == EXIF_TAG_ARTIST ||
+		tag == EXIF_TAG_SOFTWARE ||
+		tag == EXIF_TAG_COPYRIGHT ||
+		tag == EXIF_TAG_DATE_TIME ||
+		tag == EXIF_TAG_DATE_TIME_ORIGINAL ||
+		tag == EXIF_TAG_DATE_TIME_DIGITIZED ||
+		tag == EXIF_TAG_SUB_SEC_TIME ||
+		tag == EXIF_TAG_SUB_SEC_TIME_ORIGINAL ||
+		tag == EXIF_TAG_SUB_SEC_TIME_DIGITIZED );
+}
+
+static gboolean
+tag_is_utf16( ExifTag tag )
+{
+	return( tag == EXIF_TAG_XP_TITLE || 
+		tag == EXIF_TAG_XP_COMMENT || 
+		tag == EXIF_TAG_XP_AUTHOR || 
+		tag == EXIF_TAG_XP_KEYWORDS ||
+		tag == EXIF_TAG_XP_SUBJECT );
+}
+
+/* Set a libexif-formatted string entry. 
+ */
+static void
+vips_exif_alloc_string( ExifEntry *entry, unsigned long components )
+{
+	ExifMem *mem;
+
+	g_assert( !entry->data );
+
+	/* The string in the entry must be allocated with the same allocator
+	 * that was used to allocate the entry itself. We can't do this
+	 * because the allocator is private :( so we must assume the entry was
+	 * created with the default one.
+	 */
+	mem = exif_mem_new_default();
+
+	/* EXIF_FORMAT_UNDEFINED is correct for EXIF_TAG_USER_COMMENT, our 
+	 * caller should change this if it wishes.
+	 */
+	entry->data = exif_mem_alloc( mem, components );
+        entry->size = components;
+        entry->components = components;
+        entry->format = EXIF_FORMAT_UNDEFINED;
+
+	VIPS_FREEF( exif_mem_unref, mem );
+}
+
+/* The final " (xx, yy, zz, kk)" part of the string (if present) was
+ * added by us in _to_s(), we must remove it before setting the string 
+ * back again.
+ *
+ * It may not be there if the user has changed the string.
+ */
+static char *
+drop_tail( const char *data )
+{
+	char *str;
+	char *p;
+
+	str = g_strdup( data );
+
+	p = str + strlen( str );
+	if( p > str &&
+		*(p = g_utf8_prev_char( p )) == ')' &&
+		(p = g_utf8_strrchr( p, -1, (gunichar) '(')) &&
+		p > str &&
+		*(p = g_utf8_prev_char( p )) == ' ' )
+		*p = '\0';
+
+	return( str );
+}
+
+/* special header required for EXIF_TAG_USER_COMMENT.
+ */
+#define ASCII_COMMENT "ASCII\0\0\0"
+
+/* Write a libvips NULL-terminated utf-8 string into a entry tagged with a
+ * encoding. UserComment is like this, for example.
+ */
+static void
+vips_exif_set_string_encoding( ExifData *ed, 
+	ExifEntry *entry, unsigned long component, const char *data )
+{
+	char *str;
+	char *ascii;
+	int len;
+
+	str = drop_tail( data );
+
+	/* libexif can only really save ASCII to things like UserComment.
+	 */
+	ascii = g_str_to_ascii( str, NULL );
+
+	/* libexif comment strings are not NULL-terminated, and have an 
+	 * encoding tag (always ASCII) in the first 8 bytes.
+	 */
+	len = strlen( ascii );
+	vips_exif_alloc_string( entry, sizeof( ASCII_COMMENT ) - 1 + len );
+	memcpy( entry->data, ASCII_COMMENT, sizeof( ASCII_COMMENT ) - 1 );
+        memcpy( entry->data + sizeof( ASCII_COMMENT ) - 1, ascii, len );
+
+	g_free( ascii ); 
+	g_free( str );
+}
+
+/* Write a libvips NULL-terminated utf-8 string into an ASCII entry. Tags like
+ * ImageDescription work like this.
+ */
+static void
+vips_exif_set_string_ascii( ExifData *ed, 
+	ExifEntry *entry, unsigned long component, const char *data )
+{
+	char *str;
+	char *ascii;
+	int len;
+
+	str = drop_tail( data );
+	ascii = g_str_to_ascii( str, NULL );
+
+	/* ASCII strings are NULL-terminated.
+	 */
+	len = strlen( ascii );
+	vips_exif_alloc_string( entry, len + 1 );
+        memcpy( entry->data, ascii, len + 1 );
+        entry->format = EXIF_FORMAT_ASCII;
+
+	g_free( ascii ); 
+	g_free( str );
+}
+
+/* Write a libvips NULL-terminated utf-8 string into a utf16 entry.
+ */
+static void
+vips_exif_set_string_utf16( ExifData *ed, 
+	ExifEntry *entry, unsigned long component, const char *data )
+{
+	char *str;
+	gunichar2 *utf16;
+	glong len;
+
+	str = drop_tail( data );
+
+	utf16 = g_utf8_to_utf16( str, -1, NULL, &len, NULL );
+
+	/* libexif utf16 strings are NULL-terminated.
+	 */
+	vips_exif_alloc_string( entry, (len + 1) * 2 );
+	memcpy( entry->data, utf16, (len + 1) * 2 ); 
+        entry->format = EXIF_FORMAT_BYTE;
+
+	g_free( utf16 ); 
+	g_free( str );
+}
+
 /* Write a tag. Update what's there, or make a new one.
 */
 static void
@ -722,12 +918,22 @@ vips_exif_set_tag( ExifData *ed, int ifd, ExifTag tag, write_fn fn, void *data )
 		/* tag must be set before calling exif_content_add_entry.
 		 */
 		entry->tag = tag; 
-
 		exif_content_add_entry( ed->ifd[ifd], entry );
-		exif_entry_initialize( entry, tag );
 		exif_entry_unref( entry );

-		fn( ed, entry, 0, data );
+		/* libexif makes us have a special path for string-valued
+		 * fields :(
+		 */
+		if( tag_is_encoding( tag ) ) 
+			vips_exif_set_string_encoding( ed, entry, 0, data );
+		else if( tag_is_ascii( tag ) ) 
+			vips_exif_set_string_ascii( ed, entry, 0, data );
+		else if( tag_is_utf16( tag ) )
+			vips_exif_set_string_utf16( ed, entry, 0, data );
+		else {
+			exif_entry_initialize( entry, tag );
+			fn( ed, entry, 0, data );
+		}
 	}
 }

@ -869,53 +1075,73 @@ vips_exif_set_thumbnail( ExifData *ed, VipsImage *im )
 	return( 0 );
 }

-/* See also vips_exif_to_s() ... keep in sync.
+/* Skip any spaces.
+ */
+static const char *
+skip_space( const char *p )
+{
+	while( p && *p == ' ' )
+		p += 1;
+
+	return( p );
+}
+
+/* Skip to the end of this non-space sequence.
+ */
+static const char *
+skip_nonspace( const char *p )
+{
+	while( p && *p && *p != ' ' )
+		p += 1;
+
+	return( p );
+}
+
+/* See also vips_exif_to_s() ... keep in sync. Only the numeric types are
+ * handled here, since they can be updated. For string types, we have to
+ * destroy and recreate, see above. 
 */
 static void
 vips_exif_from_s( ExifData *ed, ExifEntry *entry, const char *value )
 {
 	unsigned long i;
 	const char *p;
+	int v;

-	if( entry->format != EXIF_FORMAT_SHORT &&
-		entry->format != EXIF_FORMAT_SSHORT &&
-		entry->format != EXIF_FORMAT_LONG &&
-		entry->format != EXIF_FORMAT_SLONG &&
-		entry->format != EXIF_FORMAT_RATIONAL &&
-		entry->format != EXIF_FORMAT_SRATIONAL )
-		return;
-	if( entry->components >= 10 )
-		return;
+	if( entry->format == EXIF_FORMAT_SHORT ||
+		entry->format == EXIF_FORMAT_SSHORT ||
+		entry->format == EXIF_FORMAT_LONG ||
+		entry->format == EXIF_FORMAT_SLONG ) {
+		if( entry->components >= 10 )
+			return;

-	/* Skip any leading spaces.
-	 */
-	p = value;
-	while( *p == ' ' )
-		p += 1;
+		p = value;
+		for( i = 0; i < entry->components; i++ ) {
+			if( !(p = skip_space( p )) )
+			       break;	

-	for( i = 0; i < entry->components; i++ ) {
-		if( entry->format == EXIF_FORMAT_SHORT || 
-			entry->format == EXIF_FORMAT_SSHORT || 
-			entry->format == EXIF_FORMAT_LONG || 
-			entry->format == EXIF_FORMAT_SLONG ) {
-			int value = atof( p );
+			v = atof( p );
+			vips_exif_set_int( ed, entry, i, &v );

-			vips_exif_set_int( ed, entry, i, &value );
+			p = skip_nonspace( p );
 		}
-		else if( entry->format == EXIF_FORMAT_RATIONAL || 
-			entry->format == EXIF_FORMAT_SRATIONAL ) 
+	}
+	else if( entry->format == EXIF_FORMAT_RATIONAL ||
+		entry->format == EXIF_FORMAT_SRATIONAL ) {
+		if( entry->components >= 10 )
+			return;
+
+		p = value;
+		for( i = 0; i < entry->components; i++ ) {
+			if( !(p = skip_space( p )) )
+			       break;	
+
 			vips_exif_set_rational( ed, entry, i, (void *) p );

-		/* Skip to the next set of spaces, then to the beginning of
-		 * the next item.
-		 */
-		while( *p && *p != ' ' )
-			p += 1;
-		while( *p == ' ' )
-			p += 1;
-		if( !*p )
-			break;
+			p = skip_nonspace( p );
+		}
 	}
+
 }

 static void 
@ -982,7 +1208,7 @@ vips_exif_exif_entry( ExifEntry *entry, VipsExifRemove *ve )
 	char vips_name_txt[256];
 	VipsBuf vips_name = VIPS_BUF_STATIC( vips_name_txt );

-	if( !(tag_name = exif_tag_get_name( entry->tag )) )
+	if( !(tag_name = vips_exif_entry_get_name( entry )) )
 		return;

 	vips_buf_appendf( &vips_name, "exif-ifd%d-%s", 
@ -1001,6 +1227,14 @@ vips_exif_exif_entry( ExifEntry *entry, VipsExifRemove *ve )
 	if( strcmp( tag_name, "Orientation" ) == 0 &&
 		vips_image_get_typeof( ve->image, VIPS_META_ORIENTATION ) )
 		ve->to_remove = g_slist_prepend( ve->to_remove, entry );
+
+	/* If this is a string tag, we must also remove it ready for
+	 * recreation, see the comment below.
+	 */
+	if( tag_is_encoding( entry->tag ) ||
+		tag_is_ascii( entry->tag ) ||
+		tag_is_utf16( entry->tag ) )
+		ve->to_remove = g_slist_prepend( ve->to_remove, entry );
 }

 static void *
@ -1012,7 +1246,7 @@ vips_exif_exif_remove( ExifEntry *entry, VipsExifRemove *ve )
 	char vips_name_txt[256];
 	VipsBuf vips_name = VIPS_BUF_STATIC( vips_name_txt );

-	tag_name = exif_tag_get_name( entry->tag );
+	tag_name = vips_exif_entry_get_name( entry );
 	vips_buf_appendf( &vips_name, "exif-ifd%d-%s", 
 		exif_entry_get_ifd( entry ), tag_name );

@ -1044,14 +1278,17 @@ vips_exif_update( ExifData *ed, VipsImage *image )

 	VIPS_DEBUG_MSG( "vips_exif_update: \n" );

-	/* Walk the image and add any exif- that's set in image metadata.
-	 */
-	vips_image_map( image, vips_exif_image_field, ed );
-
 	/* If this exif came from the image (rather than being an exif block we
 	 * have made afresh), then any fields which are in the block but not on
 	 * the image must have been deliberately removed. Remove them from the
 	 * block as well.
+	 *
+	 * If there are any string-valued fields (eg. comment etc.) which
+	 * exist as libvips metadata tags, we must also remove those from the
+	 * exif block.
+	 *
+	 * libexif does not allow you to change string lengths, you must make
+	 * new tags, so we have to remove ready to re-add.
 	 */
 	if( vips_image_get_typeof( image, VIPS_META_EXIF_NAME ) ) {
 		ve.image = image;
@ -1060,6 +1297,10 @@ vips_exif_update( ExifData *ed, VipsImage *image )
 			(ExifDataForeachContentFunc) vips_exif_exif_content, 
 			&ve );
 	}
+
+	/* Walk the image and add any exif- that's set in image metadata.
+	 */
+	vips_image_map( image, vips_exif_image_field, ed );
 }

 /* Examine the metadata tags on the image and update the EXIF block.
--- a/libvips/foreign/vips2tiff.c
+++ b/libvips/foreign/vips2tiff.c
@ -176,6 +176,8 @@
 * 24/10/17
 * 	- no error on page-height not a factor of image height, just don't
 * 	  write multipage
+ * 2/7/18
+ * 	- copy EXTRASAMPLES to pyramid layers
 */

 /*
@ -563,7 +565,9 @@ wtiff_write_header( Wtiff *wtiff, Layer *layer )
 	if( wtiff->compression == COMPRESSION_JPEG ) 
 		TIFFSetField( tif, TIFFTAG_JPEGQUALITY, wtiff->jpqual );

-	if( wtiff->predictor != VIPS_FOREIGN_TIFF_PREDICTOR_NONE ) 
+	if( (wtiff->compression == VIPS_FOREIGN_TIFF_COMPRESSION_DEFLATE ||
+		wtiff->compression == VIPS_FOREIGN_TIFF_COMPRESSION_LZW) &&
+		wtiff->predictor != VIPS_FOREIGN_TIFF_PREDICTOR_NONE ) 
 		TIFFSetField( tif, TIFFTAG_PREDICTOR, wtiff->predictor );

 	/* Don't write mad resolutions (eg. zero), it confuses some programs.
@ -1566,6 +1570,7 @@ wtiff_copy_tiff( Wtiff *wtiff, TIFF *out, TIFF *in )
 	tdata_t buf;
 	ttile_t tile;
 	ttile_t n;
+	uint16 *a;

 	/* All the fields we might have set.
 	 */
@ -1586,8 +1591,8 @@ wtiff_copy_tiff( Wtiff *wtiff, TIFF *out, TIFF *in )
 	CopyField( TIFFTAG_ROWSPERSTRIP, i32 );
 	CopyField( TIFFTAG_SUBFILETYPE, i32 );

-	if( wtiff->predictor != VIPS_FOREIGN_TIFF_PREDICTOR_NONE ) 
-		TIFFSetField( out, TIFFTAG_PREDICTOR, wtiff->predictor );
+	if( TIFFGetField( in, TIFFTAG_EXTRASAMPLES, &i16, &a ) ) 
+		TIFFSetField( out, TIFFTAG_EXTRASAMPLES, i16, a );

 	/* TIFFTAG_JPEGQUALITY is a pesudo-tag, so we can't copy it.
 	 * Set explicitly from Wtiff.
--- a/m4/gif.m4
+++ b/m4/gif.m4
@ -40,9 +40,9 @@ if test "$GIFLIB_INCLUDES" = ""; then
  ], [
    # gif_lib.h is not in the standard search path, try
    # $prefix
-    giflib_save_INCLUDES="$INCLUDES"
+    giflib_save_CFLAGS="$CFLAGS"

-    INCLUDES="-I${prefix}/include $INCLUDES"
+    CFLAGS="-I${prefix}/include $CFLAGS"

    AC_TRY_COMPILE([#include <gif_lib.h>],[int a;],[
      GIFLIB_INCLUDES="-I${prefix}/include"
@ -50,17 +50,17 @@ if test "$GIFLIB_INCLUDES" = ""; then
      GIFLIB_INCLUDES="no"
    ])

-    INCLUDES=$giflib_save_INCLUDES
+    CFLAGS=$giflib_save_CFLAGS
  ])
 fi

 # Now for the libraries
 if test "$GIFLIB_LIBS" = ""; then
  giflib_save_LIBS="$LIBS"
-  giflib_save_INCLUDES="$INCLUDES"
+  giflib_save_CFLAGS="$CFLAGS"

  LIBS="-lgif $LIBS"
-  INCLUDES="$GIFLIB_INCLUDES $INCLUDES"
+  CFLAGS="$GIFLIB_INCLUDES $CFLAGS"

  # Try the standard search path first
  AC_TRY_LINK([#include <gif_lib.h>],[DGifSlurp(0)], [
@ -78,7 +78,7 @@ if test "$GIFLIB_LIBS" = ""; then
  ])

  LIBS="$giflib_save_LIBS"
-  INCLUDES="$giflib_save_INCLUDES"
+  CFLAGS="$giflib_save_CFLAGS"
 fi

 AC_SUBST(GIFLIB_LIBS)
--- a/m4/jpeg.m4
+++ b/m4/jpeg.m4
@ -41,9 +41,9 @@ if test "$JPEG_INCLUDES" = ""; then
  ], [
    # jpeglib.h is not in the standard search path, try
    # $prefix
-    jpeg_save_INCLUDES="$INCLUDES"
+    jpeg_save_CFLAGS="$CFLAGS"

-    INCLUDES="-I${prefix}/include $INCLUDES"
+    CFLAGS="-I${prefix}/include $CFLAGS"

    AC_TRY_COMPILE([#include <stdio.h>
      #include <jpeglib.h>],[int a;],[
@ -52,17 +52,17 @@ if test "$JPEG_INCLUDES" = ""; then
      JPEG_INCLUDES="no"
    ])

-    INCLUDES=$jpeg_save_INCLUDES
+    CFLAGS=$jpeg_save_CFLAGS
  ])
 fi

 # Now for the libraries
 if test "$JPEG_LIBS" = ""; then
  jpeg_save_LIBS="$LIBS"
-  jpeg_save_INCLUDES="$INCLUDES"
+  jpeg_save_CFLAGS="$CFLAGS"

  LIBS="-ljpeg $LIBS"
-  INCLUDES="$JPEG_INCLUDES $INCLUDES"
+  CFLAGS="$JPEG_INCLUDES $CFLAGS"

  # Try the standard search path first
  AC_TRY_LINK([#include <stdio.h>
@ -84,7 +84,7 @@ if test "$JPEG_LIBS" = ""; then
  ])

  LIBS="$jpeg_save_LIBS"
-  INCLUDES="$jpeg_save_INCLUDES"
+  CFLAGS="$jpeg_save_CFLAGS"
 fi

 AC_SUBST(JPEG_LIBS)
--- a/m4/png.m4
+++ b/m4/png.m4
@ -40,9 +40,9 @@ if test "$PNG_INCLUDES" = ""; then
  ], [
    # png.h is not in the standard search path, try
    # $prefix
-    png_save_INCLUDES="$INCLUDES"
+    png_save_CFLAGS="$CFLAGS"

-    INCLUDES="-I${prefix}/include $INCLUDES"
+    CFLAGS="-I${prefix}/include $CFLAGS"

    AC_TRY_COMPILE([#include <png.h>],[int a;],[
      PNG_INCLUDES="-I${prefix}/include"
@ -50,17 +50,17 @@ if test "$PNG_INCLUDES" = ""; then
      PNG_INCLUDES="no"
    ])

-    INCLUDES=$png_save_INCLUDES
+    CFLAGS=$png_save_CFLAGS
  ])
 fi

 # Now for the libraries
 if test "$PNG_LIBS" = ""; then
  png_save_LIBS="$LIBS"
-  png_save_INCLUDES="$INCLUDES"
+  png_save_CFLAGS="$CFLAGS"

  LIBS="-lpng $LIBS"
-  INCLUDES="$PNG_INCLUDES $INCLUDES"
+  CFLAGS="$PNG_INCLUDES $CFLAGS"

  # Try the standard search path first
  AC_TRY_LINK([#include <png.h>],[png_access_version_number()], [
@ -78,7 +78,7 @@ if test "$PNG_LIBS" = ""; then
  ])

  LIBS="$png_save_LIBS"
-  INCLUDES="$png_save_INCLUDES"
+  CFLAGS="$png_save_CFLAGS"
 fi

 AC_SUBST(PNG_LIBS)
--- a/m4/tiff.m4
+++ b/m4/tiff.m4
@ -40,9 +40,9 @@ if test "$TIFF_INCLUDES" = ""; then
  ], [
    # tiff.h is not in the standard search path, try
    # $prefix
-    tiff_save_INCLUDES="$INCLUDES"
+    tiff_save_CFLAGS="$CFLAGS"

-    INCLUDES="-I${prefix}/include $INCLUDES"
+    CFLAGS="-I${prefix}/include $CFLAGS"

    AC_TRY_COMPILE([#include <tiff.h>],[int a;],[
      TIFF_INCLUDES="-I${prefix}/include"
@ -50,17 +50,17 @@ if test "$TIFF_INCLUDES" = ""; then
      TIFF_INCLUDES="no"
    ])

-    INCLUDES=$tiff_save_INCLUDES
+    CFLAGS=$tiff_save_CFLAGS
  ])
 fi

 # Now for the libraries
 if test "$TIFF_LIBS" = ""; then
  tiff_save_LIBS="$LIBS"
-  tiff_save_INCLUDES="$INCLUDES"
+  tiff_save_CFLAGS="$CFLAGS"

  LIBS="-ltiff -lm $LIBS"
-  INCLUDES="$TIFF_INCLUDES $INCLUDES"
+  CFLAGS="$TIFF_INCLUDES $CFLAGS"

  # Try the standard search path first
  AC_TRY_LINK([#include <tiff.h>],[TIFFGetVersion()], [
@ -78,7 +78,7 @@ if test "$TIFF_LIBS" = ""; then
  ])

  LIBS="$tiff_save_LIBS"
-  INCLUDES="$tiff_save_INCLUDES"
+  CFLAGS="$tiff_save_CFLAGS"
 fi

 AC_SUBST(TIFF_LIBS)
--- a/m4/webp.m4
+++ b/m4/webp.m4
@ -41,9 +41,9 @@ if test "$LIBWEBP_INCLUDES" = ""; then
  ], [
    # webp/decode.h is not in the standard search path, try
    # $prefix
-    libwebp_save_INCLUDES="$INCLUDES"
+    libwebp_save_CFLAGS="$CFLAGS"

-    INCLUDES="-I${prefix}/include $INCLUDES"
+    CFLAGS="-I${prefix}/include $CFLAGS"

    AC_TRY_COMPILE([#include <webp/decode.h>],[int a;],[
      LIBWEBP_INCLUDES="-I${prefix}/include"
@ -51,17 +51,17 @@ if test "$LIBWEBP_INCLUDES" = ""; then
      LIBWEBP_INCLUDES="no"
    ])

-    INCLUDES=$libwebp_save_INCLUDES
+    CFLAGS=$libwebp_save_CFLAGS
  ])
 fi

 # Now for the libraries
 if test "$LIBWEBP_LIBS" = ""; then
  libwebp_save_LIBS="$LIBS"
-  libwebp_save_INCLUDES="$INCLUDES"
+  libwebp_save_CFLAGS="$CFLAGS"

  LIBS="-lwebp $LIBS"
-  INCLUDES="$LIBWEBP_INCLUDES $INCLUDES"
+  CFLAGS="$LIBWEBP_INCLUDES $CFLAGS"

  # Try the standard search path first
  AC_TRY_LINK([#include <webp/decode.h>],[WebPInitDecoderConfig(0)], [
@ -79,7 +79,7 @@ if test "$LIBWEBP_LIBS" = ""; then
  ])

  LIBS="$libwebp_save_LIBS"
-  INCLUDES="$libwebp_save_INCLUDES"
+  CFLAGS="$libwebp_save_CFLAGS"
 fi

 AC_SUBST(LIBWEBP_LIBS)
--- a/m4/zlib.m4
+++ b/m4/zlib.m4
@ -42,9 +42,9 @@ if test "$ZLIB_INCLUDES" = ""; then
  ], [
    # zlib.h is not in the standard search path, try
    # $prefix
-    zlib_save_INCLUDES="$INCLUDES"
+    zlib_save_CFLAGS="$CFLAGS"

-    INCLUDES="-I${prefix}/include $INCLUDES"
+    CFLAGS="-I${prefix}/include $CFLAGS"

    AC_TRY_COMPILE([#include <stdio.h>
      #include <zlib.h>],[int a;],[
@ -53,17 +53,17 @@ if test "$ZLIB_INCLUDES" = ""; then
      ZLIB_INCLUDES="no"
    ])

-    INCLUDES=$zlib_save_INCLUDES
+    CFLAGS=$zlib_save_CFLAGS
  ])
 fi

 # Now for the libraries
 if test "$ZLIB_LIBS" = ""; then
  zlib_save_LIBS="$LIBS"
-  zlib_save_INCLUDES="$INCLUDES"
+  zlib_save_CFLAGS="$CFLAGS"

  LIBS="-lz $LIBS"
-  INCLUDES="$ZLIB_INCLUDES $INCLUDES"
+  CFLAGS="$ZLIB_INCLUDES $CFLAGS"

  # Try the standard search path first
  AC_TRY_LINK([#include <stdio.h>
@ -85,7 +85,7 @@ if test "$ZLIB_LIBS" = ""; then
  ])

  LIBS="$zlib_save_LIBS"
-  INCLUDES="$zlib_save_INCLUDES"
+  CFLAGS="$zlib_save_CFLAGS"
 fi

 AC_SUBST(ZLIB_LIBS)
--- a/test/README.md
+++ b/test/README.md
@ -0,0 +1,5 @@
+# libvips test suite
+
+This is in two parts: a few simple bash scripts in this directory are run on
+"make check", and a fancier Python test suite that's run by Travis on each
+commit.
--- a/test/images/Landscape_6.jpg
+++ b/test/images/Landscape_6.jpg
--- a/test/images/avg152T1_LR_nifti.nii.gz
+++ b/test/images/avg152T1_LR_nifti.nii.gz
--- a/test/test-suite/init.py
+++ b/test/test-suite/init.py
--- a/test/test-suite/conftest.py
+++ b/test/test-suite/conftest.py
@ -0,0 +1,4 @@
+import sys
+import os
+
+sys.path.append(os.path.join(os.path.dirname(__file__), 'helpers'))
--- a/test/test-suite/helpers/helpers.py
+++ b/test/test-suite/helpers/helpers.py
@ -0,0 +1,232 @@
+# vim: set fileencoding=utf-8 :
+# test helpers
+import os
+import tempfile
+import pytest
+
+import pyvips
+
+IMAGES = os.path.join(os.path.dirname(__file__), os.pardir, 'images')
+JPEG_FILE = os.path.join(IMAGES, "йцук.jpg")
+SRGB_FILE = os.path.join(IMAGES, "sRGB.icm")
+MATLAB_FILE = os.path.join(IMAGES, "sample.mat")
+PNG_FILE = os.path.join(IMAGES, "sample.png")
+TIF_FILE = os.path.join(IMAGES, "sample.tif")
+OME_FILE = os.path.join(IMAGES, "multi-channel-z-series.ome.tif")
+ANALYZE_FILE = os.path.join(IMAGES, "t00740_tr1_segm.hdr")
+GIF_FILE = os.path.join(IMAGES, "cramps.gif")
+WEBP_FILE = os.path.join(IMAGES, "1.webp")
+EXR_FILE = os.path.join(IMAGES, "sample.exr")
+FITS_FILE = os.path.join(IMAGES, "WFPC2u5780205r_c0fx.fits")
+OPENSLIDE_FILE = os.path.join(IMAGES, "CMU-1-Small-Region.svs")
+PDF_FILE = os.path.join(IMAGES, "ISO_12233-reschart.pdf")
+CMYK_PDF_FILE = os.path.join(IMAGES, "cmyktest.pdf")
+SVG_FILE = os.path.join(IMAGES, "vips-profile.svg")
+SVGZ_FILE = os.path.join(IMAGES, "vips-profile.svgz")
+SVG_GZ_FILE = os.path.join(IMAGES, "vips-profile.svg.gz")
+GIF_ANIM_FILE = os.path.join(IMAGES, "cogs.gif")
+DICOM_FILE = os.path.join(IMAGES, "dicom_test_image.dcm")
+BMP_FILE = os.path.join(IMAGES, "MARBLES.BMP")
+
+unsigned_formats = [pyvips.BandFormat.UCHAR,
+                    pyvips.BandFormat.USHORT,
+                    pyvips.BandFormat.UINT]
+signed_formats = [pyvips.BandFormat.CHAR,
+                  pyvips.BandFormat.SHORT,
+                  pyvips.BandFormat.INT]
+float_formats = [pyvips.BandFormat.FLOAT,
+                 pyvips.BandFormat.DOUBLE]
+complex_formats = [pyvips.BandFormat.COMPLEX,
+                   pyvips.BandFormat.DPCOMPLEX]
+int_formats = unsigned_formats + signed_formats
+noncomplex_formats = int_formats + float_formats
+all_formats = int_formats + float_formats + complex_formats
+
+colour_colourspaces = [pyvips.Interpretation.XYZ,
+                       pyvips.Interpretation.LAB,
+                       pyvips.Interpretation.LCH,
+                       pyvips.Interpretation.CMC,
+                       pyvips.Interpretation.LABS,
+                       pyvips.Interpretation.SCRGB,
+                       pyvips.Interpretation.HSV,
+                       pyvips.Interpretation.SRGB,
+                       pyvips.Interpretation.YXY]
+coded_colourspaces = [pyvips.Interpretation.LABQ]
+mono_colourspaces = [pyvips.Interpretation.B_W]
+sixteenbit_colourspaces = [pyvips.Interpretation.GREY16,
+                           pyvips.Interpretation.RGB16]
+all_colourspaces = colour_colourspaces + mono_colourspaces + \
+                   coded_colourspaces + sixteenbit_colourspaces
+
+max_value = {pyvips.BandFormat.UCHAR: 0xff,
+             pyvips.BandFormat.USHORT: 0xffff,
+             pyvips.BandFormat.UINT: 0xffffffff,
+             pyvips.BandFormat.CHAR: 0x7f,
+             pyvips.BandFormat.SHORT: 0x7fff,
+             pyvips.BandFormat.INT: 0x7fffffff,
+             pyvips.BandFormat.FLOAT: 1.0,
+             pyvips.BandFormat.DOUBLE: 1.0,
+             pyvips.BandFormat.COMPLEX: 1.0,
+             pyvips.BandFormat.DPCOMPLEX: 1.0}
+
+sizeof_format = {pyvips.BandFormat.UCHAR: 1,
+                 pyvips.BandFormat.USHORT: 2,
+                 pyvips.BandFormat.UINT: 4,
+                 pyvips.BandFormat.CHAR: 1,
+                 pyvips.BandFormat.SHORT: 2,
+                 pyvips.BandFormat.INT: 4,
+                 pyvips.BandFormat.FLOAT: 4,
+                 pyvips.BandFormat.DOUBLE: 8,
+                 pyvips.BandFormat.COMPLEX: 8,
+                 pyvips.BandFormat.DPCOMPLEX: 16}
+
+rot45_angles = [pyvips.Angle45.D0,
+                pyvips.Angle45.D45,
+                pyvips.Angle45.D90,
+                pyvips.Angle45.D135,
+                pyvips.Angle45.D180,
+                pyvips.Angle45.D225,
+                pyvips.Angle45.D270,
+                pyvips.Angle45.D315]
+
+rot45_angle_bonds = [pyvips.Angle45.D0,
+                     pyvips.Angle45.D315,
+                     pyvips.Angle45.D270,
+                     pyvips.Angle45.D225,
+                     pyvips.Angle45.D180,
+                     pyvips.Angle45.D135,
+                     pyvips.Angle45.D90,
+                     pyvips.Angle45.D45]
+
+rot_angles = [pyvips.Angle.D0,
+              pyvips.Angle.D90,
+              pyvips.Angle.D180,
+              pyvips.Angle.D270]
+
+rot_angle_bonds = [pyvips.Angle.D0,
+                   pyvips.Angle.D270,
+                   pyvips.Angle.D180,
+                   pyvips.Angle.D90]
+
+
+# an expanding zip ... if either of the args is a scalar or a one-element list,
+# duplicate it down the other side
+def zip_expand(x, y):
+    # handle singleton list case
+    if isinstance(x, list) and len(x) == 1:
+        x = x[0]
+    if isinstance(y, list) and len(y) == 1:
+        y = y[0]
+
+    if isinstance(x, list) and isinstance(y, list):
+        return list(zip(x, y))
+    elif isinstance(x, list):
+        return [[i, y] for i in x]
+    elif isinstance(y, list):
+        return [[x, j] for j in y]
+    else:
+        return [[x, y]]
+
+
+# run a 1-ary function on a thing -- loop over elements if the
+# thing is a list
+def run_fn(fn, x):
+    if isinstance(x, list):
+        return [fn(i) for i in x]
+    else:
+        return fn(x)
+
+
+# make a temp filename with the specified suffix and in the
+# specified directory
+def temp_filename(directory, suffix):
+    temp_name = next(tempfile._get_candidate_names())
+    filename = os.path.join(directory, temp_name + suffix)
+
+    return filename
+
+
+# test for an operator exists
+def have(name):
+    return pyvips.type_find("VipsOperation", name) != 0
+
+
+def skip_if_no(operation_name):
+    return pytest.mark.skipif(not have(operation_name),
+                        reason='no {}, skipping test'.format(operation_name))
+
+
+# run a 2-ary function on two things -- loop over elements pairwise if the
+# things are lists
+def run_fn2(fn, x, y):
+    if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+        return fn(x, y)
+    elif isinstance(x, list) or isinstance(y, list):
+        return [fn(i, j) for i, j in zip_expand(x, y)]
+    else:
+        return fn(x, y)
+
+
+# test a pair of things which can be lists for approx. equality
+def assert_almost_equal_objects(a, b, threshold=0.0001, msg=''):
+    # print 'assertAlmostEqualObjects %s = %s' % (a, b)
+    assert all([pytest.approx(x, abs=threshold) == y
+                for x, y in zip_expand(a, b)]), msg
+
+
+# test a pair of things which can be lists for equality
+def assert_equal_objects(a, b, msg=''):
+    # print 'assertEqualObjects %s = %s' % (a, b)
+    assert all([x == y for x, y in zip_expand(a, b)]), msg
+
+
+# test a pair of things which can be lists for difference less than a
+# threshold
+def assert_less_threshold(a, b, diff):
+    assert all([abs(x - y) < diff for x, y in zip_expand(a, b)])
+
+
+# run a function on an image and on a single pixel, the results
+# should match
+def run_cmp(message, im, x, y, fn):
+    a = im(x, y)
+    v1 = fn(a)
+    im2 = fn(im)
+    v2 = im2(x, y)
+    assert_almost_equal_objects(v1, v2, msg=message)
+
+
+# run a function on an image,
+# 50,50 and 10,10 should have different values on the test image
+def run_image(message, im, fn):
+    run_cmp(message, im, 50, 50, fn)
+    run_cmp(message, im, 10, 10, fn)
+
+
+# run a function on (image, constant), and on (constant, image).
+# 50,50 and 10,10 should have different values on the test image
+def run_const(message, fn, im, c):
+    run_cmp(message, im, 50, 50, lambda x: run_fn2(fn, x, c))
+    run_cmp(message, im, 50, 50, lambda x: run_fn2(fn, c, x))
+    run_cmp(message, im, 10, 10, lambda x: run_fn2(fn, x, c))
+    run_cmp(message, im, 10, 10, lambda x: run_fn2(fn, c, x))
+
+
+# run a function on a pair of images and on a pair of pixels, the results
+# should match
+def run_cmp2(message, left, right, x, y, fn):
+    a = left(x, y)
+    b = right(x, y)
+    v1 = fn(a, b)
+    after = fn(left, right)
+    v2 = after(x, y)
+    assert_almost_equal_objects(v1, v2, msg=message)
+
+
+# run a function on a pair of images
+# 50,50 and 10,10 should have different values on the test image
+def run_image2(message, left, right, fn):
+    run_cmp2(message, left, right, 50, 50,
+             lambda x, y: run_fn2(fn, x, y))
+    run_cmp2(message, left, right, 10, 10,
+             lambda x, y: run_fn2(fn, x, y))
--- a/test/test-suite/images/1.webp
+++ b/test/test-suite/images/1.webp
--- a/test/test-suite/images/CMU-1-Small-Region.svs
+++ b/test/test-suite/images/CMU-1-Small-Region.svs
--- a/test/test-suite/images/ISO_12233-reschart.pdf
+++ b/test/test-suite/images/ISO_12233-reschart.pdf
--- a/test/test-suite/images/MARBLES.BMP
+++ b/test/test-suite/images/MARBLES.BMP
--- a/test/test-suite/images/WFPC2u5780205r_c0fx.fits
+++ b/test/test-suite/images/WFPC2u5780205r_c0fx.fits
--- a/test/test-suite/images/blankpage.pdf
+++ b/test/test-suite/images/blankpage.pdf
--- a/test/test-suite/images/blankpage.pdf.png
+++ b/test/test-suite/images/blankpage.pdf.png
--- a/test/test-suite/images/blankpage.svg
+++ b/test/test-suite/images/blankpage.svg
--- a/test/test-suite/images/blankpage.svg.png
+++ b/test/test-suite/images/blankpage.svg.png
--- a/test/test-suite/images/cmyktest.pdf
+++ b/test/test-suite/images/cmyktest.pdf
--- a/test/test-suite/images/cogs.gif
+++ b/test/test-suite/images/cogs.gif
--- a/test/test-suite/images/cramps.gif
+++ b/test/test-suite/images/cramps.gif
--- a/test/test-suite/images/dicom_test_image.dcm
+++ b/test/test-suite/images/dicom_test_image.dcm
--- a/test/test-suite/images/multi-channel-z-series.ome.tif
+++ b/test/test-suite/images/multi-channel-z-series.ome.tif
--- a/test/test-suite/images/sRGB.icm
+++ b/test/test-suite/images/sRGB.icm
--- a/test/test-suite/images/sample.exr
+++ b/test/test-suite/images/sample.exr
--- a/test/test-suite/images/sample.mat
+++ b/test/test-suite/images/sample.mat
--- a/test/test-suite/images/sample.png
+++ b/test/test-suite/images/sample.png
--- a/test/test-suite/images/sample.tif
+++ b/test/test-suite/images/sample.tif
--- a/test/test-suite/images/t00740_tr1_segm.hdr
+++ b/test/test-suite/images/t00740_tr1_segm.hdr
--- a/test/test-suite/images/t00740_tr1_segm.img
+++ b/test/test-suite/images/t00740_tr1_segm.img
--- a/test/test-suite/images/trans-x.gif
+++ b/test/test-suite/images/trans-x.gif
--- a/test/test-suite/images/trans-x.png
+++ b/test/test-suite/images/trans-x.png
--- a/test/test-suite/images/vips-profile.svg
+++ b/test/test-suite/images/vips-profile.svg
--- a/test/test-suite/images/vips-profile.svg.gz
+++ b/test/test-suite/images/vips-profile.svg.gz
--- a/test/test-suite/images/vips-profile.svgz
+++ b/test/test-suite/images/vips-profile.svgz
--- a/test/test-suite/images/йцук.jpg
+++ b/test/test-suite/images/йцук.jpg
--- a/test/test-suite/test_arithmetic.py
+++ b/test/test-suite/test_arithmetic.py
@ -0,0 +1,641 @@
+# vim: set fileencoding=utf-8 :
+import math
+import pytest
+
+import pyvips
+from helpers import unsigned_formats, float_formats, noncomplex_formats, \
+    all_formats, run_fn, run_image2, run_const, run_cmp, \
+    assert_almost_equal_objects
+
+
+class TestArithmetic:
+    def run_arith(self, fn, fmt=all_formats):
+        [run_image2(fn.__name__ + ' image', x.cast(y), x.cast(z), fn)
+         for x in self.all_images for y in fmt for z in fmt]
+
+    def run_arith_const(self, fn, fmt=all_formats):
+        [run_const(fn.__name__ + ' scalar', fn, x.cast(y), 2)
+         for x in self.all_images for y in fmt]
+        [run_const(fn.__name__ + ' vector', fn, self.colour.cast(y),
+                   [1, 2, 3])
+         for y in fmt]
+
+    # run a function on an image,
+    # 50,50 and 10,10 should have different values on the test image
+    def run_imageunary(self, message, im, fn):
+        run_cmp(message, im, 50, 50, lambda x: run_fn(fn, x))
+        run_cmp(message, im, 10, 10, lambda x: run_fn(fn, x))
+
+    def run_unary(self, images, fn, fmt=all_formats):
+        [self.run_imageunary(fn.__name__ + ' image', x.cast(y), fn)
+         for x in images for y in fmt]
+
+    @classmethod
+    def setup_class(cls):
+        im = pyvips.Image.mask_ideal(100, 100, 0.5,
+                                     reject=True, optical=True)
+        cls.colour = im * [1, 2, 3] + [2, 3, 4]
+        cls.mono = cls.colour.extract_band(1)
+        cls.all_images = [cls.mono, cls.colour]
+
+    # test all operator overloads we define
+
+    def test_add(self):
+        def add(x, y):
+            return x + y
+
+        self.run_arith_const(add)
+        self.run_arith(add)
+
+    def test_sub(self):
+        def sub(x, y):
+            return x - y
+
+        self.run_arith_const(sub)
+        self.run_arith(sub)
+
+    def test_mul(self):
+        def mul(x, y):
+            return x * y
+
+        self.run_arith_const(mul)
+        self.run_arith(mul)
+
+    def test_div(self):
+        def div(x, y):
+            return x / y
+
+        # (const / image) needs (image ** -1), which won't work for complex
+        self.run_arith_const(div, fmt=noncomplex_formats)
+        self.run_arith(div)
+
+    def test_floordiv(self):
+        def my_floordiv(x, y):
+            return x // y
+
+        # (const // image) needs (image ** -1), which won't work for complex
+        self.run_arith_const(my_floordiv, fmt=noncomplex_formats)
+        self.run_arith(my_floordiv, fmt=noncomplex_formats)
+
+    def test_pow(self):
+        def my_pow(x, y):
+            return x ** y
+
+        # (image ** x) won't work for complex images ... just test non-complex
+        self.run_arith_const(my_pow, fmt=noncomplex_formats)
+        self.run_arith(my_pow, fmt=noncomplex_formats)
+
+    def test_and(self):
+        def my_and(x, y):
+            # python doesn't allow bools on float
+            if isinstance(x, float):
+                x = int(x)
+            if isinstance(y, float):
+                y = int(y)
+            return x & y
+
+        self.run_arith_const(my_and, fmt=noncomplex_formats)
+        self.run_arith(my_and, fmt=noncomplex_formats)
+
+    def test_or(self):
+        def my_or(x, y):
+            # python doesn't allow bools on float
+            if isinstance(x, float):
+                x = int(x)
+            if isinstance(y, float):
+                y = int(y)
+            return x | y
+
+        self.run_arith_const(my_or, fmt=noncomplex_formats)
+        self.run_arith(my_or, fmt=noncomplex_formats)
+
+    def test_xor(self):
+        def my_xor(x, y):
+            # python doesn't allow bools on float
+            if isinstance(x, float):
+                x = int(x)
+            if isinstance(y, float):
+                y = int(y)
+            return x ^ y
+
+        self.run_arith_const(my_xor, fmt=noncomplex_formats)
+        self.run_arith(my_xor, fmt=noncomplex_formats)
+
+    def test_more(self):
+        def more(x, y):
+            if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+                return x > y
+            else:
+                if x > y:
+                    return 255
+                else:
+                    return 0
+
+        self.run_arith_const(more)
+        self.run_arith(more)
+
+    def test_moreeq(self):
+        def moreeq(x, y):
+            if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+                return x >= y
+            else:
+                if x >= y:
+                    return 255
+                else:
+                    return 0
+
+        self.run_arith_const(moreeq)
+        self.run_arith(moreeq)
+
+    def test_less(self):
+        def less(x, y):
+            if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+                return x < y
+            else:
+                if x < y:
+                    return 255
+                else:
+                    return 0
+
+        self.run_arith_const(less)
+        self.run_arith(less)
+
+    def test_lesseq(self):
+        def lesseq(x, y):
+            if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+                return x <= y
+            else:
+                if x <= y:
+                    return 255
+                else:
+                    return 0
+
+        self.run_arith_const(lesseq)
+        self.run_arith(lesseq)
+
+    def test_equal(self):
+        def equal(x, y):
+            if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+                return x == y
+            else:
+                if x == y:
+                    return 255
+                else:
+                    return 0
+
+        self.run_arith_const(equal)
+        self.run_arith(equal)
+
+    def test_noteq(self):
+        def noteq(x, y):
+            if isinstance(x, pyvips.Image) or isinstance(y, pyvips.Image):
+                return x != y
+            else:
+                if x != y:
+                    return 255
+                else:
+                    return 0
+
+        self.run_arith_const(noteq)
+        self.run_arith(noteq)
+
+    def test_abs(self):
+        def my_abs(x):
+            return abs(x)
+
+        im = -self.colour
+        self.run_unary([im], my_abs)
+
+    def test_lshift(self):
+        def my_lshift(x):
+            # python doesn't allow float << int
+            if isinstance(x, float):
+                x = int(x)
+            return x << 2
+
+        # we don't support constant << image, treat as a unary
+        self.run_unary(self.all_images, my_lshift, fmt=noncomplex_formats)
+
+    def test_rshift(self):
+        def my_rshift(x):
+            # python doesn't allow float >> int
+            if isinstance(x, float):
+                x = int(x)
+            return x >> 2
+
+        # we don't support constant >> image, treat as a unary
+        self.run_unary(self.all_images, my_rshift, fmt=noncomplex_formats)
+
+    def test_mod(self):
+        def my_mod(x):
+            return x % 2
+
+        # we don't support constant % image, treat as a unary
+        self.run_unary(self.all_images, my_mod, fmt=noncomplex_formats)
+
+    def test_pos(self):
+        def my_pos(x):
+            return +x
+
+        self.run_unary(self.all_images, my_pos)
+
+    def test_neg(self):
+        def my_neg(x):
+            return -x
+
+        self.run_unary(self.all_images, my_neg)
+
+    def test_invert(self):
+        def my_invert(x):
+            if isinstance(x, float):
+                x = int(x)
+            return ~x & 0xff
+
+        # ~image is trimmed to image max so it's hard to test for all formats
+        # just test uchar
+        self.run_unary(self.all_images, my_invert,
+                       fmt=[pyvips.BandFormat.UCHAR])
+
+    # test the rest of VipsArithmetic
+
+    def test_avg(self):
+        im = pyvips.Image.black(50, 100)
+        test = im.insert(im + 100, 50, 0, expand=True)
+
+        for fmt in all_formats:
+            assert pytest.approx(test.cast(fmt).avg()) == 50
+
+    def test_deviate(self):
+        im = pyvips.Image.black(50, 100)
+        test = im.insert(im + 100, 50, 0, expand=True)
+
+        for fmt in noncomplex_formats:
+            assert pytest.approx(test.cast(fmt).deviate(), abs=0.01) == 50
+
+    def test_polar(self):
+        im = pyvips.Image.black(100, 100) + 100
+        im = im.complexform(im)
+
+        im = im.polar()
+
+        assert pytest.approx(im.real().avg()) == 100 * 2 ** 0.5
+        assert pytest.approx(im.imag().avg()) == 45
+
+    def test_rect(self):
+        im = pyvips.Image.black(100, 100)
+        im = (im + 100 * 2 ** 0.5).complexform(im + 45)
+
+        im = im.rect()
+
+        assert pytest.approx(im.real().avg()) == 100
+        assert pytest.approx(im.imag().avg()) == 100
+
+    def test_conjugate(self):
+        im = pyvips.Image.black(100, 100) + 100
+        im = im.complexform(im)
+
+        im = im.conj()
+
+        assert pytest.approx(im.real().avg()) == 100
+        assert pytest.approx(im.imag().avg()) == -100
+
+    def test_histfind(self):
+        im = pyvips.Image.black(50, 100)
+        test = im.insert(im + 10, 50, 0, expand=True)
+
+        for fmt in all_formats:
+            hist = test.cast(fmt).hist_find()
+            assert_almost_equal_objects(hist(0, 0), [5000])
+            assert_almost_equal_objects(hist(10, 0), [5000])
+            assert_almost_equal_objects(hist(5, 0), [0])
+
+        test = test * [1, 2, 3]
+
+        for fmt in all_formats:
+            hist = test.cast(fmt).hist_find(band=0)
+            assert_almost_equal_objects(hist(0, 0), [5000])
+            assert_almost_equal_objects(hist(10, 0), [5000])
+            assert_almost_equal_objects(hist(5, 0), [0])
+
+            hist = test.cast(fmt).hist_find(band=1)
+            assert_almost_equal_objects(hist(0, 0), [5000])
+            assert_almost_equal_objects(hist(20, 0), [5000])
+            assert_almost_equal_objects(hist(5, 0), [0])
+
+    def test_histfind_indexed(self):
+        im = pyvips.Image.black(50, 100)
+        test = im.insert(im + 10, 50, 0, expand=True)
+        index = test // 10
+
+        for x in noncomplex_formats:
+            for y in [pyvips.BandFormat.UCHAR, pyvips.BandFormat.USHORT]:
+                a = test.cast(x)
+                b = index.cast(y)
+                hist = a.hist_find_indexed(b)
+
+                assert_almost_equal_objects(hist(0, 0), [0])
+                assert_almost_equal_objects(hist(1, 0), [50000])
+
+    def test_histfind_ndim(self):
+        im = pyvips.Image.black(100, 100) + [1, 2, 3]
+
+        for fmt in noncomplex_formats:
+            hist = im.cast(fmt).hist_find_ndim()
+
+            assert_almost_equal_objects(hist(0, 0)[0], 10000)
+            assert_almost_equal_objects(hist(5, 5)[5], 0)
+
+            hist = im.cast(fmt).hist_find_ndim(bins=1)
+
+            assert_almost_equal_objects(hist(0, 0)[0], 10000)
+            assert hist.width == 1
+            assert hist.height == 1
+            assert hist.bands == 1
+
+    def test_hough_circle(self):
+        test = pyvips.Image.black(100, 100).draw_circle(100, 50, 50, 40)
+
+        for fmt in all_formats:
+            im = test.cast(fmt)
+            hough = im.hough_circle(min_radius=35, max_radius=45)
+
+            v, x, y = hough.maxpos()
+            vec = hough(x, y)
+            r = vec.index(v) + 35
+
+            assert pytest.approx(x) == 50
+            assert pytest.approx(y) == 50
+            assert pytest.approx(r) == 40
+
+    @pytest.mark.skipif(not pyvips.base.at_least_libvips(8, 7),
+                        reason="requires libvips >= 8.7")
+    def test_hough_line(self):
+        # hough_line changed the way it codes parameter space in 8.7 ... don't
+        # test earlier versions
+        test = pyvips.Image.black(100, 100).draw_line(100, 10, 90, 90, 10)
+
+        for fmt in all_formats:
+            im = test.cast(fmt)
+            hough = im.hough_line()
+
+            v, x, y = hough.maxpos()
+
+            angle = 180.0 * x // hough.width
+            distance = test.height * y // hough.height
+
+            assert pytest.approx(angle) == 45
+            assert pytest.approx(distance) == 70
+
+    def test_sin(self):
+        def my_sin(x):
+            if isinstance(x, pyvips.Image):
+                return x.sin()
+            else:
+                return math.sin(math.radians(x))
+
+        self.run_unary(self.all_images, my_sin, fmt=noncomplex_formats)
+
+    def test_cos(self):
+        def my_cos(x):
+            if isinstance(x, pyvips.Image):
+                return x.cos()
+            else:
+                return math.cos(math.radians(x))
+
+        self.run_unary(self.all_images, my_cos, fmt=noncomplex_formats)
+
+    def test_tan(self):
+        def my_tan(x):
+            if isinstance(x, pyvips.Image):
+                return x.tan()
+            else:
+                return math.tan(math.radians(x))
+
+        self.run_unary(self.all_images, my_tan, fmt=noncomplex_formats)
+
+    def test_asin(self):
+        def my_asin(x):
+            if isinstance(x, pyvips.Image):
+                return x.asin()
+            else:
+                return math.degrees(math.asin(x))
+
+        im = (pyvips.Image.black(100, 100) + [1, 2, 3]) / 3.0
+        self.run_unary([im], my_asin, fmt=noncomplex_formats)
+
+    def test_acos(self):
+        def my_acos(x):
+            if isinstance(x, pyvips.Image):
+                return x.acos()
+            else:
+                return math.degrees(math.acos(x))
+
+        im = (pyvips.Image.black(100, 100) + [1, 2, 3]) / 3.0
+        self.run_unary([im], my_acos, fmt=noncomplex_formats)
+
+    def test_atan(self):
+        def my_atan(x):
+            if isinstance(x, pyvips.Image):
+                return x.atan()
+            else:
+                return math.degrees(math.atan(x))
+
+        im = (pyvips.Image.black(100, 100) + [1, 2, 3]) / 3.0
+        self.run_unary([im], my_atan, fmt=noncomplex_formats)
+
+    def test_log(self):
+        def my_log(x):
+            if isinstance(x, pyvips.Image):
+                return x.log()
+            else:
+                return math.log(x)
+
+        self.run_unary(self.all_images, my_log, fmt=noncomplex_formats)
+
+    def test_log10(self):
+        def my_log10(x):
+            if isinstance(x, pyvips.Image):
+                return x.log10()
+            else:
+                return math.log10(x)
+
+        self.run_unary(self.all_images, my_log10, fmt=noncomplex_formats)
+
+    def test_exp(self):
+        def my_exp(x):
+            if isinstance(x, pyvips.Image):
+                return x.exp()
+            else:
+                return math.exp(x)
+
+        self.run_unary(self.all_images, my_exp, fmt=noncomplex_formats)
+
+    def test_exp10(self):
+        def my_exp10(x):
+            if isinstance(x, pyvips.Image):
+                return x.exp10()
+            else:
+                return math.pow(10, x)
+
+        self.run_unary(self.all_images, my_exp10, fmt=noncomplex_formats)
+
+    def test_floor(self):
+        def my_floor(x):
+            if isinstance(x, pyvips.Image):
+                return x.floor()
+            else:
+                return math.floor(x)
+
+        self.run_unary(self.all_images, my_floor)
+
+    def test_ceil(self):
+        def my_ceil(x):
+            if isinstance(x, pyvips.Image):
+                return x.ceil()
+            else:
+                return math.ceil(x)
+
+        self.run_unary(self.all_images, my_ceil)
+
+    def test_rint(self):
+        def my_rint(x):
+            if isinstance(x, pyvips.Image):
+                return x.rint()
+            else:
+                return round(x)
+
+        self.run_unary(self.all_images, my_rint)
+
+    def test_sign(self):
+        def my_sign(x):
+            if isinstance(x, pyvips.Image):
+                return x.sign()
+            else:
+                if x > 0:
+                    return 1
+                elif x < 0:
+                    return -1
+                else:
+                    return 0
+
+        self.run_unary(self.all_images, my_sign)
+
+    def test_max(self):
+        test = pyvips.Image.black(100, 100).draw_rect(100, 40, 50, 1, 1)
+
+        for fmt in all_formats:
+            v = test.cast(fmt).max()
+
+            assert pytest.approx(v) == 100
+            v, x, y = test.cast(fmt).maxpos()
+            assert pytest.approx(v) == 100
+            assert pytest.approx(x) == 40
+            assert pytest.approx(y) == 50
+
+    def test_min(self):
+        test = (pyvips.Image.black(100, 100) + 100).draw_rect(0, 40, 50, 1, 1)
+
+        for fmt in all_formats:
+            v = test.cast(fmt).min()
+
+            assert pytest.approx(v) == 0
+            v, x, y = test.cast(fmt).minpos()
+            assert pytest.approx(v) == 0
+            assert pytest.approx(x) == 40
+            assert pytest.approx(y) == 50
+
+    def test_measure(self):
+        im = pyvips.Image.black(50, 50)
+        test = im.insert(im + 10, 50, 0, expand=True)
+
+        for x in noncomplex_formats:
+            a = test.cast(x)
+            matrix = a.measure(2, 1)
+            [p1] = matrix(0, 0)
+            [p2] = matrix(0, 1)
+
+            assert pytest.approx(p1) == 0
+            assert pytest.approx(p2) == 10
+
+    def test_find_trim(self):
+        if pyvips.type_find("VipsOperation", "find_trim") != 0:
+            im = pyvips.Image.black(50, 60) + 100
+            test = im.embed(10, 20, 200, 300, extend="white")
+
+            for x in unsigned_formats + float_formats:
+                a = test.cast(x)
+                left, top, width, height = a.find_trim()
+
+                assert left == 10
+                assert top == 20
+                assert width == 50
+                assert height == 60
+
+            test_rgb = test.bandjoin([test, test])
+            left, top, width, height = test_rgb.find_trim(background=[255, 255,
+                                                                      255])
+            assert left == 10
+            assert top == 20
+            assert width == 50
+            assert height == 60
+
+    def test_profile(self):
+        test = pyvips.Image.black(100, 100).draw_rect(100, 40, 50, 1, 1)
+
+        for fmt in noncomplex_formats:
+            columns, rows = test.cast(fmt).profile()
+
+            v, x, y = columns.minpos()
+            assert pytest.approx(v) == 50
+            assert pytest.approx(x) == 40
+            assert pytest.approx(y) == 0
+
+            v, x, y = rows.minpos()
+            assert pytest.approx(v) == 40
+            assert pytest.approx(x) == 0
+            assert pytest.approx(y) == 50
+
+    def test_project(self):
+        im = pyvips.Image.black(50, 50)
+        test = im.insert(im + 10, 50, 0, expand=True)
+
+        for fmt in noncomplex_formats:
+            columns, rows = test.cast(fmt).project()
+
+            assert_almost_equal_objects(columns(10, 0), [0])
+            assert_almost_equal_objects(columns(70, 0), [50 * 10])
+
+            assert_almost_equal_objects(rows(0, 10), [50 * 10])
+
+    def test_stats(self):
+        im = pyvips.Image.black(50, 50)
+        test = im.insert(im + 10, 50, 0, expand=True)
+
+        for x in noncomplex_formats:
+            a = test.cast(x)
+            matrix = a.stats()
+
+            assert_almost_equal_objects(matrix(0, 0), [a.min()])
+            assert_almost_equal_objects(matrix(1, 0), [a.max()])
+            assert_almost_equal_objects(matrix(2, 0), [50 * 50 * 10])
+            assert_almost_equal_objects(matrix(3, 0), [50 * 50 * 100])
+            assert_almost_equal_objects(matrix(4, 0), [a.avg()])
+            assert_almost_equal_objects(matrix(5, 0), [a.deviate()])
+
+            assert_almost_equal_objects(matrix(0, 1), [a.min()])
+            assert_almost_equal_objects(matrix(1, 1), [a.max()])
+            assert_almost_equal_objects(matrix(2, 1), [50 * 50 * 10])
+            assert_almost_equal_objects(matrix(3, 1), [50 * 50 * 100])
+            assert_almost_equal_objects(matrix(4, 1), [a.avg()])
+            assert_almost_equal_objects(matrix(5, 1), [a.deviate()])
+
+    def test_sum(self):
+        for fmt in all_formats:
+            im = pyvips.Image.black(50, 50)
+            im2 = [(im + x).cast(fmt) for x in range(0, 100, 10)]
+            im3 = pyvips.Image.sum(im2)
+            assert pytest.approx(im3.max()) == sum(range(0, 100, 10))
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_colour.py
+++ b/test/test-suite/test_colour.py
@ -0,0 +1,155 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+from helpers import JPEG_FILE, SRGB_FILE, colour_colourspaces, \
+    mono_colourspaces, assert_almost_equal_objects
+
+
+class TestColour:
+    def test_colourspace(self):
+        # mid-grey in Lab ... put 42 in the extra band, it should be copied
+        # unmodified
+        test = pyvips.Image.black(100, 100) + [50, 0, 0, 42]
+        test = test.copy(interpretation=pyvips.Interpretation.LAB)
+
+        # a long series should come in a circle
+        im = test
+        for col in colour_colourspaces + [pyvips.Interpretation.LAB]:
+            im = im.colourspace(col)
+            assert im.interpretation == col
+
+            for i in range(0, 4):
+                min_l = im.extract_band(i).min()
+                max_h = im.extract_band(i).max()
+                assert pytest.approx(min_l) == max_h
+
+            pixel = im(10, 10)
+            assert pytest.approx(pixel[3], 0.01) == 42
+
+        # alpha won't be equal for RGB16, but it should be preserved if we go
+        # there and back
+        im = im.colourspace(pyvips.Interpretation.RGB16)
+        im = im.colourspace(pyvips.Interpretation.LAB)
+
+        before = test(10, 10)
+        after = im(10, 10)
+        assert_almost_equal_objects(before, after, threshold=0.1)
+
+        # go between every pair of colour spaces
+        for start in colour_colourspaces:
+            for end in colour_colourspaces:
+                im = test.colourspace(start)
+                im2 = im.colourspace(end)
+                im3 = im2.colourspace(pyvips.Interpretation.LAB)
+
+                before = test(10, 10)
+                after = im3(10, 10)
+
+                assert_almost_equal_objects(before, after, threshold=0.1)
+
+        # test Lab->XYZ on mid-grey
+        # checked against http://www.brucelindbloom.com
+        im = test.colourspace(pyvips.Interpretation.XYZ)
+        after = im(10, 10)
+        assert_almost_equal_objects(after, [17.5064, 18.4187, 20.0547, 42])
+
+        # grey->colour->grey should be equal
+        for mono_fmt in mono_colourspaces:
+            test_grey = test.colourspace(mono_fmt)
+            im = test_grey
+            for col in colour_colourspaces + [mono_fmt]:
+                im = im.colourspace(col)
+                assert im.interpretation == col
+            [before, alpha_before] = test_grey(10, 10)
+            [after, alpha_after] = im(10, 10)
+            assert abs(alpha_after - alpha_before) < 1
+            if mono_fmt == pyvips.Interpretation.GREY16:
+                # GREY16 can wind up rather different due to rounding
+                assert abs(after - before) < 30
+            else:
+                # but 8-bit we should hit exactly
+                assert abs(after - before) < 1
+
+    # test results from Bruce Lindbloom's calculator:
+    # http://www.brucelindbloom.com
+    def test_dE00(self):
+        # put 42 in the extra band, it should be copied unmodified
+        reference = pyvips.Image.black(100, 100) + [50, 10, 20, 42]
+        reference = reference.copy(interpretation=pyvips.Interpretation.LAB)
+        sample = pyvips.Image.black(100, 100) + [40, -20, 10]
+        sample = sample.copy(interpretation=pyvips.Interpretation.LAB)
+
+        difference = reference.dE00(sample)
+        result, alpha = difference(10, 10)
+        assert pytest.approx(result, 0.001) == 30.238
+        assert pytest.approx(alpha, 0.001) == 42.0
+
+    def test_dE76(self):
+        # put 42 in the extra band, it should be copied unmodified
+        reference = pyvips.Image.black(100, 100) + [50, 10, 20, 42]
+        reference = reference.copy(interpretation=pyvips.Interpretation.LAB)
+        sample = pyvips.Image.black(100, 100) + [40, -20, 10]
+        sample = sample.copy(interpretation=pyvips.Interpretation.LAB)
+
+        difference = reference.dE76(sample)
+        result, alpha = difference(10, 10)
+        assert pytest.approx(result, 0.001) == 33.166
+        assert pytest.approx(alpha, 0.001) == 42.0
+
+    # the vips CMC calculation is based on distance in a colorspace
+    # derived from the CMC formula, so it won't match exactly ...
+    # see vips_LCh2CMC() for details
+    def test_dECMC(self):
+        reference = pyvips.Image.black(100, 100) + [50, 10, 20, 42]
+        reference = reference.copy(interpretation=pyvips.Interpretation.LAB)
+        sample = pyvips.Image.black(100, 100) + [55, 11, 23]
+        sample = sample.copy(interpretation=pyvips.Interpretation.LAB)
+
+        difference = reference.dECMC(sample)
+        result, alpha = difference(10, 10)
+        assert abs(result - 4.97) < 0.5
+        assert pytest.approx(alpha, 0.001) == 42.0
+
+    def test_icc(self):
+        test = pyvips.Image.new_from_file(JPEG_FILE)
+
+        im = test.icc_import().icc_export()
+        assert im.dE76(test).max() < 6
+
+        im = test.icc_import()
+        im2 = im.icc_export(depth=16)
+        assert im2.format == pyvips.BandFormat.USHORT
+        im3 = im2.icc_import()
+        assert (im - im3).abs().max() < 3
+
+        im = test.icc_import(intent=pyvips.Intent.ABSOLUTE)
+        im2 = im.icc_export(intent=pyvips.Intent.ABSOLUTE)
+        assert im2.dE76(test).max() < 6
+
+        im = test.icc_import()
+        im2 = im.icc_export(output_profile=SRGB_FILE)
+        im3 = im.colourspace(pyvips.Interpretation.SRGB)
+        assert im2.dE76(im3).max() < 6
+
+        before_profile = test.get("icc-profile-data")
+        im = test.icc_transform(SRGB_FILE)
+        after_profile = im.get("icc-profile-data")
+        im2 = test.icc_import()
+        im3 = im2.colourspace(pyvips.Interpretation.SRGB)
+        assert im.dE76(im3).max() < 6
+        assert len(before_profile) != len(after_profile)
+
+        im = test.icc_import(input_profile=SRGB_FILE)
+        im2 = test.icc_import()
+        assert 6 < im.dE76(im2).max()
+
+        im = test.icc_import(pcs=pyvips.PCS.XYZ)
+        assert im.interpretation == pyvips.Interpretation.XYZ
+
+        im = test.icc_import()
+        assert im.interpretation == pyvips.Interpretation.LAB
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_conversion.py
+++ b/test/test-suite/test_conversion.py
@ -0,0 +1,767 @@
+# vim: set fileencoding=utf-8 :
+from functools import reduce
+import pytest
+
+import pyvips
+from helpers import JPEG_FILE, unsigned_formats, \
+    signed_formats, float_formats, int_formats, \
+    noncomplex_formats, all_formats, max_value, \
+    sizeof_format, rot45_angles, rot45_angle_bonds, \
+    rot_angles, rot_angle_bonds, run_cmp, run_cmp2, \
+    assert_almost_equal_objects
+
+
+class TestConversion:
+    # run a function on an image,
+    # 50,50 and 10,10 should have different values on the test image
+    # don't loop over band elements
+    def run_image_pixels(self, message, im, fn):
+        run_cmp(message, im, 50, 50, fn)
+        run_cmp(message, im, 10, 10, fn)
+
+    # run a function on a pair of images
+    # 50,50 and 10,10 should have different values on the test image
+    # don't loop over band elements
+    def run_image_pixels2(self, message, left, right, fn):
+        run_cmp2(message, left, right, 50, 50, fn)
+        run_cmp2(message, left, right, 10, 10, fn)
+
+    def run_unary(self, images, fn, fmt=all_formats):
+        [self.run_image_pixels(fn.__name__ + (' %s' % y), x.cast(y), fn)
+         for x in images for y in fmt]
+
+    def run_binary(self, images, fn, fmt=all_formats):
+        [self.run_image_pixels2(fn.__name__ + (' %s %s' % (y, z)),
+                                x.cast(y), x.cast(z), fn)
+         for x in images for y in fmt for z in fmt]
+
+    @classmethod
+    def setup_class(cls):
+        im = pyvips.Image.mask_ideal(100, 100, 0.5,
+                                     reject=True, optical=True)
+        cls.colour = (im * [1, 2, 3] + [2, 3, 4]).copy(interpretation="srgb")
+        cls.mono = cls.colour[1].copy(interpretation="b-w")
+        cls.all_images = [cls.mono, cls.colour]
+        cls.image = pyvips.Image.jpegload(JPEG_FILE)
+
+    def test_band_and(self):
+        def band_and(x):
+            if isinstance(x, pyvips.Image):
+                return x.bandand()
+            else:
+                return [reduce(lambda a, b: int(a) & int(b), x)]
+
+        self.run_unary(self.all_images, band_and, fmt=int_formats)
+
+    def test_band_or(self):
+        def band_or(x):
+            if isinstance(x, pyvips.Image):
+                return x.bandor()
+            else:
+                return [reduce(lambda a, b: int(a) | int(b), x)]
+
+        self.run_unary(self.all_images, band_or, fmt=int_formats)
+
+    def test_band_eor(self):
+        def band_eor(x):
+            if isinstance(x, pyvips.Image):
+                return x.bandeor()
+            else:
+                return [reduce(lambda a, b: int(a) ^ int(b), x)]
+
+        self.run_unary(self.all_images, band_eor, fmt=int_formats)
+
+    def test_bandjoin(self):
+        def bandjoin(x, y):
+            if isinstance(x, pyvips.Image) and isinstance(y, pyvips.Image):
+                return x.bandjoin(y)
+            else:
+                return x + y
+
+        self.run_binary(self.all_images, bandjoin)
+
+    def test_bandjoin_const(self):
+        x = self.colour.bandjoin(1)
+        assert x.bands == 4
+        assert x[3].avg() == 1
+
+        x = self.colour.bandjoin([1, 2])
+        assert x.bands == 5
+        assert x[3].avg() == 1
+        assert x[4].avg() == 2
+
+    def test_bandmean(self):
+        def bandmean(x):
+            if isinstance(x, pyvips.Image):
+                return x.bandmean()
+            else:
+                return [sum(x) // len(x)]
+
+        self.run_unary(self.all_images, bandmean, fmt=noncomplex_formats)
+
+    def test_bandrank(self):
+        def median(x, y):
+            joined = [[a, b] for a, b in zip(x, y)]
+            # .sort() isn't a function, so we have to run this as a separate
+            # pass
+            [z.sort() for z in joined]
+            return [z[len(z) // 2] for z in joined]
+
+        def bandrank(x, y):
+            if isinstance(x, pyvips.Image) and isinstance(y, pyvips.Image):
+                return x.bandrank([y])
+            else:
+                return median(x, y)
+
+        self.run_binary(self.all_images, bandrank, fmt=noncomplex_formats)
+
+        # we can mix images and constants, and set the index arg
+        a = self.mono.bandrank([2], index=0)
+        b = (self.mono < 2).ifthenelse(self.mono, 2)
+        assert (a - b).abs().min() == 0
+
+    def test_cache(self):
+        def cache(x):
+            if isinstance(x, pyvips.Image):
+                return x.cache()
+            else:
+                return x
+
+        self.run_unary(self.all_images, cache)
+
+    def test_copy(self):
+        x = self.colour.copy(interpretation=pyvips.Interpretation.LAB)
+        assert x.interpretation == pyvips.Interpretation.LAB
+        x = self.colour.copy(xres=42)
+        assert x.xres == 42
+        x = self.colour.copy(yres=42)
+        assert x.yres == 42
+        x = self.colour.copy(xoffset=42)
+        assert x.xoffset == 42
+        x = self.colour.copy(yoffset=42)
+        assert x.yoffset == 42
+        x = self.colour.copy(coding=pyvips.Coding.NONE)
+        assert x.coding == pyvips.Coding.NONE
+
+    def test_bandfold(self):
+        x = self.mono.bandfold()
+        assert x.width == 1
+        assert x.bands == self.mono.width
+
+        y = x.bandunfold()
+        assert y.width == self.mono.width
+        assert y.bands == 1
+        assert x.avg() == y.avg()
+
+        x = self.mono.bandfold(factor=2)
+        assert x.width == self.mono.width / 2
+        assert x.bands == 2
+
+        y = x.bandunfold(factor=2)
+        assert y.width == self.mono.width
+        assert y.bands == 1
+        assert x.avg() == y.avg()
+
+    def test_byteswap(self):
+        x = self.mono.cast("ushort")
+        y = x.byteswap().byteswap()
+        assert x.width == y.width
+        assert x.height == y.height
+        assert x.bands == y.bands
+        assert x.avg() == y.avg()
+
+    def test_embed(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            im = test.embed(20, 20,
+                            self.colour.width + 40,
+                            self.colour.height + 40)
+            pixel = im(10, 10)
+            assert_almost_equal_objects(pixel, [0, 0, 0])
+            pixel = im(30, 30)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+            pixel = im(im.width - 10, im.height - 10)
+            assert_almost_equal_objects(pixel, [0, 0, 0])
+
+            im = test.embed(20, 20,
+                            self.colour.width + 40,
+                            self.colour.height + 40,
+                            extend=pyvips.Extend.COPY)
+            pixel = im(10, 10)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+            pixel = im(im.width - 10, im.height - 10)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+
+            im = test.embed(20, 20,
+                            self.colour.width + 40,
+                            self.colour.height + 40,
+                            extend=pyvips.Extend.BACKGROUND,
+                            background=[7, 8, 9])
+            pixel = im(10, 10)
+            assert_almost_equal_objects(pixel, [7, 8, 9])
+            pixel = im(im.width - 10, im.height - 10)
+            assert_almost_equal_objects(pixel, [7, 8, 9])
+
+            im = test.embed(20, 20,
+                            self.colour.width + 40,
+                            self.colour.height + 40,
+                            extend=pyvips.Extend.WHITE)
+            pixel = im(10, 10)
+            # uses 255 in all bytes of ints, 255.0 for float
+            pixel = [int(x) & 0xff for x in pixel]
+            assert_almost_equal_objects(pixel, [255, 255, 255])
+            pixel = im(im.width - 10, im.height - 10)
+            pixel = [int(x) & 0xff for x in pixel]
+            assert_almost_equal_objects(pixel, [255, 255, 255])
+
+    @pytest.mark.skipif(pyvips.type_find("VipsOperation", "gravity") == 0,
+                        reason="no gravity in this vips, skipping test")
+    def test_gravity(self):
+        im = pyvips.Image.black(1, 1) + 255
+
+        positions = [
+            ['centre', 1, 1],
+            ['north', 1, 0],
+            ['south', 1, 2],
+            ['east', 2, 1],
+            ['west', 0, 1],
+            ['north-east', 2, 0],
+            ['south-east', 2, 2],
+            ['south-west', 0, 2],
+            ['north-west', 0, 0]
+        ]
+
+        for direction, x, y in positions:
+            im2 = im.gravity(direction, 3, 3)
+            assert_almost_equal_objects(im2(x, y), [255])
+            assert_almost_equal_objects(im2.avg(), 255.0 / 9.0)
+
+    def test_extract(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            pixel = test(30, 30)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+
+            sub = test.extract_area(25, 25, 10, 10)
+
+            pixel = sub(5, 5)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+
+            sub = test.extract_band(1, n=2)
+
+            pixel = sub(30, 30)
+            assert_almost_equal_objects(pixel, [3, 4])
+
+    def test_slice(self):
+        test = self.colour
+        bands = [x.avg() for x in test]
+
+        x = test[0].avg()
+        assert x == bands[0]
+
+        x = test[-1].avg()
+        assert_almost_equal_objects(x, bands[2])
+
+        x = [i.avg() for i in test[1:3]]
+        assert_almost_equal_objects(x, bands[1:3])
+
+        x = [i.avg() for i in test[1:-1]]
+        assert_almost_equal_objects(x, bands[1:-1])
+
+        x = [i.avg() for i in test[:2]]
+        assert_almost_equal_objects(x, bands[:2])
+
+        x = [i.avg() for i in test[1:]]
+        assert_almost_equal_objects(x, bands[1:])
+
+        x = [i.avg() for i in test[-1]]
+        assert_almost_equal_objects(x, bands[-1])
+
+    def test_crop(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            pixel = test(30, 30)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+
+            sub = test.crop(25, 25, 10, 10)
+
+            pixel = sub(5, 5)
+            assert_almost_equal_objects(pixel, [2, 3, 4])
+
+    @pytest.mark.skipif(pyvips.type_find("VipsOperation", "smartcrop") == 0,
+                        reason="no smartcrop, skipping test")
+    def test_smartcrop(self):
+        test = self.image.smartcrop(100, 100)
+        assert test.width == 100
+        assert test.height == 100
+
+    def test_falsecolour(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            im = test.falsecolour()
+
+            assert im.width == test.width
+            assert im.height == test.height
+            assert im.bands == 3
+
+            pixel = im(30, 30)
+            assert_almost_equal_objects(pixel, [20, 0, 41])
+
+    def test_flatten(self):
+        for fmt in unsigned_formats + [pyvips.BandFormat.SHORT,
+                                       pyvips.BandFormat.INT] + float_formats:
+            mx = 255
+            alpha = mx / 2.0
+            nalpha = mx - alpha
+            test = self.colour.bandjoin(alpha).cast(fmt)
+            pixel = test(30, 30)
+
+            predict = [int(x) * alpha / mx for x in pixel[:-1]]
+
+            im = test.flatten()
+
+            assert im.bands == 3
+            pixel = im(30, 30)
+            for x, y in zip(pixel, predict):
+                # we use float arithetic for int and uint, so the rounding
+                # differs ... don't require huge accuracy
+                assert abs(x - y) < 2
+
+            im = test.flatten(background=[100, 100, 100])
+
+            pixel = test(30, 30)
+            predict = [int(x) * alpha / mx + (100 * nalpha) / mx
+                       for x in pixel[:-1]]
+
+            assert im.bands == 3
+            pixel = im(30, 30)
+            for x, y in zip(pixel, predict):
+                assert abs(x - y) < 2
+
+    def test_premultiply(self):
+        for fmt in unsigned_formats + [pyvips.BandFormat.SHORT,
+                                       pyvips.BandFormat.INT] + float_formats:
+            mx = 255
+            alpha = mx / 2.0
+            test = self.colour.bandjoin(alpha).cast(fmt)
+            pixel = test(30, 30)
+
+            predict = [int(x) * alpha / mx for x in pixel[:-1]] + [alpha]
+
+            im = test.premultiply()
+
+            assert im.bands == test.bands
+            pixel = im(30, 30)
+            for x, y in zip(pixel, predict):
+                # we use float arithetic for int and uint, so the rounding
+                # differs ... don't require huge accuracy
+                assert abs(x - y) < 2
+
+    @pytest.mark.skipif(pyvips.type_find("VipsConversion", "composite") == 0,
+                        reason="no composite support, skipping test")
+    def test_composite(self):
+        # 50% transparent image
+        overlay = self.colour.bandjoin(128)
+        base = self.colour + 100
+        comp = base.composite(overlay, "over")
+
+        assert_almost_equal_objects(comp(0, 0), [51.8, 52.8, 53.8, 255],
+                                    threshold=0.1)
+
+    def test_unpremultiply(self):
+        for fmt in unsigned_formats + [pyvips.BandFormat.SHORT,
+                                       pyvips.BandFormat.INT] + float_formats:
+            mx = 255
+            alpha = mx / 2.0
+            test = self.colour.bandjoin(alpha).cast(fmt)
+            pixel = test(30, 30)
+
+            predict = [int(x) / (alpha / mx) for x in pixel[:-1]] + [alpha]
+
+            im = test.unpremultiply()
+
+            assert im.bands == test.bands
+            pixel = im(30, 30)
+            for x, y in zip(pixel, predict):
+                # we use float arithetic for int and uint, so the rounding
+                # differs ... don't require huge accuracy
+                assert abs(x - y) < 2
+
+    def test_flip(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            result = test.fliphor()
+            result = result.flipver()
+            result = result.fliphor()
+            result = result.flipver()
+
+            diff = (test - result).abs().max()
+
+            assert diff == 0
+
+    def test_gamma(self):
+        exponent = 2.4
+        for fmt in noncomplex_formats:
+            mx = max_value[fmt]
+            test = (self.colour + mx / 2.0).cast(fmt)
+
+            norm = mx ** exponent / mx
+            result = test.gamma()
+            before = test(30, 30)
+            after = result(30, 30)
+            predict = [x ** exponent / norm for x in before]
+            for a, b in zip(after, predict):
+                # ie. less than 1% error, rounding on 7-bit images
+                # means this is all we can expect
+                assert abs(a - b) < mx / 100.0
+
+        exponent = 1.2
+        for fmt in noncomplex_formats:
+            mx = max_value[fmt]
+            test = (self.colour + mx / 2.0).cast(fmt)
+
+            norm = mx ** exponent / mx
+            result = test.gamma(exponent=1.0 / 1.2)
+            before = test(30, 30)
+            after = result(30, 30)
+            predict = [x ** exponent / norm for x in before]
+            for a, b in zip(after, predict):
+                # ie. less than 1% error, rounding on 7-bit images
+                # means this is all we can expect
+                assert abs(a - b) < mx / 100.0
+
+    def test_grid(self):
+        test = self.colour.replicate(1, 12)
+        assert test.width == self.colour.width
+        assert test.height == self.colour.height * 12
+
+        for fmt in all_formats:
+            im = test.cast(fmt)
+            result = im.grid(test.width, 3, 4)
+            assert result.width == self.colour.width * 3
+            assert result.height == self.colour.height * 4
+
+            before = im(10, 10)
+            after = result(10 + test.width * 2, 10 + test.width * 2)
+            assert_almost_equal_objects(before, after)
+
+            before = im(50, 50)
+            after = result(50 + test.width * 2, 50 + test.width * 2)
+            assert_almost_equal_objects(before, after)
+
+    def test_ifthenelse(self):
+        test = self.mono > 3
+        for x in all_formats:
+            for y in all_formats:
+                t = (self.colour + 10).cast(x)
+                e = self.colour.cast(y)
+                r = test.ifthenelse(t, e)
+
+                assert r.width == self.colour.width
+                assert r.height == self.colour.height
+                assert r.bands == self.colour.bands
+
+                predict = e(10, 10)
+                result = r(10, 10)
+                assert_almost_equal_objects(result, predict)
+
+                predict = t(50, 50)
+                result = r(50, 50)
+                assert_almost_equal_objects(result, predict)
+
+        test = self.colour > 3
+        for x in all_formats:
+            for y in all_formats:
+                t = (self.mono + 10).cast(x)
+                e = self.mono.cast(y)
+                r = test.ifthenelse(t, e)
+
+                assert r.width == self.colour.width
+                assert r.height == self.colour.height
+                assert r.bands == self.colour.bands
+
+                cp = test(10, 10)
+                tp = t(10, 10) * 3
+                ep = e(10, 10) * 3
+                predict = [te if ce != 0 else ee
+                           for ce, te, ee in zip(cp, tp, ep)]
+                result = r(10, 10)
+                assert_almost_equal_objects(result, predict)
+
+                cp = test(50, 50)
+                tp = t(50, 50) * 3
+                ep = e(50, 50) * 3
+                predict = [te if ce != 0 else ee
+                           for ce, te, ee in zip(cp, tp, ep)]
+                result = r(50, 50)
+                assert_almost_equal_objects(result, predict)
+
+        test = self.colour > 3
+        for x in all_formats:
+            for y in all_formats:
+                t = (self.mono + 10).cast(x)
+                e = self.mono.cast(y)
+                r = test.ifthenelse(t, e, blend=True)
+
+                assert r.width == self.colour.width
+                assert r.height == self.colour.height
+                assert r.bands == self.colour.bands
+
+                result = r(10, 10)
+                assert_almost_equal_objects(result, [3, 3, 13])
+
+        test = self.mono > 3
+        r = test.ifthenelse([1, 2, 3], self.colour)
+        assert r.width == self.colour.width
+        assert r.height == self.colour.height
+        assert r.bands == self.colour.bands
+        assert r.format == self.colour.format
+        assert r.interpretation == self.colour.interpretation
+        result = r(10, 10)
+        assert_almost_equal_objects(result, [2, 3, 4])
+        result = r(50, 50)
+        assert_almost_equal_objects(result, [1, 2, 3])
+
+        test = self.mono
+        r = test.ifthenelse([1, 2, 3], self.colour, blend=True)
+        assert r.width == self.colour.width
+        assert r.height == self.colour.height
+        assert r.bands == self.colour.bands
+        assert r.format == self.colour.format
+        assert r.interpretation == self.colour.interpretation
+        result = r(10, 10)
+        assert_almost_equal_objects(result, [2, 3, 4], threshold=0.1)
+        result = r(50, 50)
+        assert_almost_equal_objects(result, [3.0, 4.9, 6.9], threshold=0.1)
+
+    def test_insert(self):
+        for x in all_formats:
+            for y in all_formats:
+                main = self.mono.cast(x)
+                sub = self.colour.cast(y)
+                r = main.insert(sub, 10, 10)
+
+                assert r.width == main.width
+                assert r.height == main.height
+                assert r.bands == sub.bands
+
+                a = r(10, 10)
+                b = sub(0, 0)
+                assert_almost_equal_objects(a, b)
+
+                a = r(0, 0)
+                b = main(0, 0) * 3
+                assert_almost_equal_objects(a, b)
+
+        for x in all_formats:
+            for y in all_formats:
+                main = self.mono.cast(x)
+                sub = self.colour.cast(y)
+                r = main.insert(sub, 10, 10, expand=True, background=100)
+
+                assert r.width == main.width + 10
+                assert r.height == main.height + 10
+                assert r.bands == sub.bands
+
+                a = r(r.width - 5, 5)
+                assert_almost_equal_objects(a, [100, 100, 100])
+
+    def test_arrayjoin(self):
+        max_width = 0
+        max_height = 0
+        max_bands = 0
+        for image in self.all_images:
+            if image.width > max_width:
+                max_width = image.width
+            if image.height > max_height:
+                max_height = image.height
+            if image.bands > max_bands:
+                max_bands = image.bands
+
+        im = pyvips.Image.arrayjoin(self.all_images)
+        assert im.width == max_width * len(self.all_images)
+        assert im.height == max_height
+        assert im.bands == max_bands
+
+        im = pyvips.Image.arrayjoin(self.all_images, across=1)
+        assert im.width == max_width
+        assert im.height == max_height * len(self.all_images)
+        assert im.bands == max_bands
+
+        im = pyvips.Image.arrayjoin(self.all_images, shim=10)
+        assert im.width == max_width * len(self.all_images) + 10 * (len(self.all_images) - 1)  # noqa: E501
+        assert im.height == max_height
+        assert im.bands == max_bands
+
+    def test_msb(self):
+        for fmt in unsigned_formats:
+            mx = max_value[fmt]
+            size = sizeof_format[fmt]
+            test = (self.colour + mx / 8.0).cast(fmt)
+            im = test.msb()
+
+            before = test(10, 10)
+            predict = [int(x) >> ((size - 1) * 8) for x in before]
+            result = im(10, 10)
+            assert_almost_equal_objects(result, predict)
+
+            before = test(50, 50)
+            predict = [int(x) >> ((size - 1) * 8) for x in before]
+            result = im(50, 50)
+            assert_almost_equal_objects(result, predict)
+
+        for fmt in signed_formats:
+            mx = max_value[fmt]
+            size = sizeof_format[fmt]
+            test = (self.colour + mx / 8.0).cast(fmt)
+            im = test.msb()
+
+            before = test(10, 10)
+            predict = [128 + (int(x) >> ((size - 1) * 8)) for x in before]
+            result = im(10, 10)
+            assert_almost_equal_objects(result, predict)
+
+            before = test(50, 50)
+            predict = [128 + (int(x) >> ((size - 1) * 8)) for x in before]
+            result = im(50, 50)
+            assert_almost_equal_objects(result, predict)
+
+        for fmt in unsigned_formats:
+            mx = max_value[fmt]
+            size = sizeof_format[fmt]
+            test = (self.colour + mx / 8.0).cast(fmt)
+            im = test.msb(band=1)
+
+            before = [test(10, 10)[1]]
+            predict = [int(x) >> ((size - 1) * 8) for x in before]
+            result = im(10, 10)
+            assert_almost_equal_objects(result, predict)
+
+            before = [test(50, 50)[1]]
+            predict = [int(x) >> ((size - 1) * 8) for x in before]
+            result = im(50, 50)
+            assert_almost_equal_objects(result, predict)
+
+    def test_recomb(self):
+        array = [[0.2, 0.5, 0.3]]
+
+        def recomb(x):
+            if isinstance(x, pyvips.Image):
+                return x.recomb(array)
+            else:
+                sum = 0
+                for i, c in zip(array[0], x):
+                    sum += i * c
+                return [sum]
+
+        self.run_unary([self.colour], recomb, fmt=noncomplex_formats)
+
+    def test_replicate(self):
+        for fmt in all_formats:
+            im = self.colour.cast(fmt)
+
+            test = im.replicate(10, 10)
+            assert test.width == self.colour.width * 10
+            assert test.height == self.colour.height * 10
+
+            before = im(10, 10)
+            after = test(10 + im.width * 2, 10 + im.width * 2)
+            assert_almost_equal_objects(before, after)
+
+            before = im(50, 50)
+            after = test(50 + im.width * 2, 50 + im.width * 2)
+            assert_almost_equal_objects(before, after)
+
+    def test_rot45(self):
+        # test has a quarter-circle in the bottom right
+        test = self.colour.crop(0, 0, 51, 51)
+        for fmt in all_formats:
+            im = test.cast(fmt)
+
+            im2 = im.rot45()
+            before = im(50, 50)
+            after = im2(25, 50)
+            assert_almost_equal_objects(before, after)
+
+            for a, b in zip(rot45_angles, rot45_angle_bonds):
+                im2 = im.rot45(angle=a)
+                after = im2.rot45(angle=b)
+                diff = (after - im).abs().max()
+                assert diff == 0
+
+    def test_rot(self):
+        # test has a quarter-circle in the bottom right
+        test = self.colour.crop(0, 0, 51, 51)
+        for fmt in all_formats:
+            im = test.cast(fmt)
+
+            im2 = im.rot(pyvips.Angle.D90)
+            before = im(50, 50)
+            after = im2(0, 50)
+            assert_almost_equal_objects(before, after)
+
+            for a, b in zip(rot_angles, rot_angle_bonds):
+                im2 = im.rot(a)
+                after = im2.rot(b)
+                diff = (after - im).abs().max()
+                assert diff == 0
+
+    def test_scaleimage(self):
+        for fmt in noncomplex_formats:
+            test = self.colour.cast(fmt)
+
+            im = test.scaleimage()
+            assert im.max() == 255
+            assert im.min() == 0
+
+            im = test.scaleimage(log=True)
+            assert im.max() == 255
+
+    def test_subsample(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            im = test.subsample(3, 3)
+            assert im.width == test.width // 3
+            assert im.height == test.height // 3
+
+            before = test(60, 60)
+            after = im(20, 20)
+            assert_almost_equal_objects(before, after)
+
+    def test_zoom(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            im = test.zoom(3, 3)
+            assert im.width == test.width * 3
+            assert im.height == test.height * 3
+
+            before = test(50, 50)
+            after = im(150, 150)
+            assert_almost_equal_objects(before, after)
+
+    def test_wrap(self):
+        for fmt in all_formats:
+            test = self.colour.cast(fmt)
+
+            im = test.wrap()
+            assert im.width == test.width
+            assert im.height == test.height
+
+            before = test(0, 0)
+            after = im(50, 50)
+            assert_almost_equal_objects(before, after)
+
+            before = test(50, 50)
+            after = im(0, 0)
+            assert_almost_equal_objects(before, after)
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_convolution.py
+++ b/test/test-suite/test_convolution.py
@ -0,0 +1,206 @@
+# vim: set fileencoding=utf-8 :
+import operator
+import pytest
+from functools import reduce
+
+import pyvips
+from helpers import noncomplex_formats, run_fn2, run_fn, \
+    assert_almost_equal_objects, assert_less_threshold
+
+
+# point convolution
+def conv(image, mask, x_position, y_position):
+    s = 0.0
+    for x in range(0, mask.width):
+        for y in range(0, mask.height):
+            m = mask(x, y)
+            i = image(x + x_position, y + y_position)
+            p = run_fn2(operator.mul, m, i)
+            s = run_fn2(operator.add, s, p)
+
+    return run_fn2(operator.truediv, s, mask.scale)
+
+
+def compass(image, mask, x_position, y_position, n_rot, fn):
+    acc = []
+    for i in range(0, n_rot):
+        result = conv(image, mask, x_position, y_position)
+        result = run_fn(abs, result)
+        acc.append(result)
+        mask = mask.rot45()
+
+    return reduce(lambda a, b: run_fn2(fn, a, b), acc)
+
+
+class TestConvolution:
+    @classmethod
+    def setup_class(cls):
+        im = pyvips.Image.mask_ideal(100, 100, 0.5, reject=True, optical=True)
+        cls.colour = im * [1, 2, 3] + [2, 3, 4]
+        cls.colour = cls.colour.copy(interpretation=pyvips.Interpretation.SRGB)
+        cls.mono = cls.colour.extract_band(1)
+        cls.mono = cls.mono.copy(interpretation=pyvips.Interpretation.B_W)
+        cls.all_images = [cls.mono, cls.colour]
+        cls.sharp = pyvips.Image.new_from_array([[-1, -1, -1],
+                                                 [-1, 16, -1],
+                                                 [-1, -1, -1]], scale=8)
+        cls.blur = pyvips.Image.new_from_array([[1, 1, 1],
+                                                [1, 1, 1],
+                                                [1, 1, 1]], scale=9)
+        cls.line = pyvips.Image.new_from_array([[1, 1, 1],
+                                                [-2, -2, -2],
+                                                [1, 1, 1]])
+        cls.sobel = pyvips.Image.new_from_array([[1, 2, 1],
+                                                 [0, 0, 0],
+                                                 [-1, -2, -1]])
+        cls.all_masks = [cls.sharp, cls.blur, cls.line, cls.sobel]
+
+    def test_conv(self):
+        for im in self.all_images:
+            for msk in self.all_masks:
+                for prec in [pyvips.Precision.INTEGER, pyvips.Precision.FLOAT]:
+                    convolved = im.conv(msk, precision=prec)
+
+                    result = convolved(25, 50)
+                    true = conv(im, msk, 24, 49)
+                    assert_almost_equal_objects(result, true)
+
+                    result = convolved(50, 50)
+                    true = conv(im, msk, 49, 49)
+                    assert_almost_equal_objects(result, true)
+
+    # don't test conva, it's still not done
+    def dont_est_conva(self):
+        for im in self.all_images:
+            for msk in self.all_masks:
+                print("msk:")
+                msk.matrixprint()
+                print("im.bands = %s" % im.bands)
+
+                convolved = im.conv(msk,
+                                    precision=pyvips.Precision.APPROXIMATE)
+
+                result = convolved(25, 50)
+                true = conv(im, msk, 24, 49)
+                print("result = %s, true = %s" % (result, true))
+                assert_less_threshold(result, true, 5)
+
+                result = convolved(50, 50)
+                true = conv(im, msk, 49, 49)
+                print("result = %s, true = %s" % (result, true))
+                assert_less_threshold(result, true, 5)
+
+    def test_compass(self):
+        for im in self.all_images:
+            for msk in self.all_masks:
+                for prec in [pyvips.Precision.INTEGER, pyvips.Precision.FLOAT]:
+                    for times in range(1, 4):
+                        convolved = im.compass(msk,
+                                               times=times,
+                                               angle=pyvips.Angle45.D45,
+                                               combine=pyvips.Combine.MAX,
+                                               precision=prec)
+
+                        result = convolved(25, 50)
+                        true = compass(im, msk, 24, 49, times, max)
+                        assert_almost_equal_objects(result, true)
+
+        for im in self.all_images:
+            for msk in self.all_masks:
+                for prec in [pyvips.Precision.INTEGER, pyvips.Precision.FLOAT]:
+                    for times in range(1, 4):
+                        convolved = im.compass(msk,
+                                               times=times,
+                                               angle=pyvips.Angle45.D45,
+                                               combine=pyvips.Combine.SUM,
+                                               precision=prec)
+
+                        result = convolved(25, 50)
+                        true = compass(im, msk, 24, 49, times, operator.add)
+                        assert_almost_equal_objects(result, true)
+
+    def test_convsep(self):
+        for im in self.all_images:
+            for prec in [pyvips.Precision.INTEGER, pyvips.Precision.FLOAT]:
+                gmask = pyvips.Image.gaussmat(2, 0.1,
+                                              precision=prec)
+                gmask_sep = pyvips.Image.gaussmat(2, 0.1,
+                                                  separable=True,
+                                                  precision=prec)
+
+                assert gmask.width == gmask.height
+                assert gmask_sep.width == gmask.width
+                assert gmask_sep.height == 1
+
+                a = im.conv(gmask, precision=prec)
+                b = im.convsep(gmask_sep, precision=prec)
+
+                a_point = a(25, 50)
+                b_point = b(25, 50)
+
+                assert_almost_equal_objects(a_point, b_point, threshold=0.1)
+
+    def test_fastcor(self):
+        for im in self.all_images:
+            for fmt in noncomplex_formats:
+                small = im.crop(20, 45, 10, 10).cast(fmt)
+                cor = im.fastcor(small)
+                v, x, y = cor.minpos()
+
+                assert v == 0
+                assert x == 25
+                assert y == 50
+
+    def test_spcor(self):
+        for im in self.all_images:
+            for fmt in noncomplex_formats:
+                small = im.crop(20, 45, 10, 10).cast(fmt)
+                cor = im.spcor(small)
+                v, x, y = cor.maxpos()
+
+                assert v == 1.0
+                assert x == 25
+                assert y == 50
+
+    def test_gaussblur(self):
+        for im in self.all_images:
+            for prec in [pyvips.Precision.INTEGER, pyvips.Precision.FLOAT]:
+                for i in range(5, 10):
+                    sigma = i / 5.0
+                    gmask = pyvips.Image.gaussmat(sigma, 0.2,
+                                                  precision=prec)
+
+                    a = im.conv(gmask, precision=prec)
+                    b = im.gaussblur(sigma, min_ampl=0.2, precision=prec)
+
+                    a_point = a(25, 50)
+                    b_point = b(25, 50)
+
+                    assert_almost_equal_objects(a_point, b_point,
+                                                threshold=0.1)
+
+    def test_sharpen(self):
+        for im in self.all_images:
+            for fmt in noncomplex_formats:
+                # old vipses used "radius", check that that still works
+                sharp = im.sharpen(radius=5)
+
+                for sigma in [0.5, 1, 1.5, 2]:
+                    im = im.cast(fmt)
+                    sharp = im.sharpen(sigma=sigma)
+
+                    # hard to test much more than this
+                    assert im.width == sharp.width
+                    assert im.height == sharp.height
+
+                    # if m1 and m2 are zero, sharpen should do nothing
+                    sharp = im.sharpen(sigma=sigma, m1=0, m2=0)
+                    sharp = sharp.colourspace(im.interpretation)
+                    # print("testing sig = %g" % sigma)
+                    # print("testing fmt = %s" % fmt)
+                    # print("max diff = %g" % (im - sharp).abs().max())
+                    assert (im - sharp).abs().max() == 0
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_create.py
+++ b/test/test-suite/test_create.py
@ -0,0 +1,439 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+from helpers import assert_almost_equal_objects
+
+
+class TestCreate:
+    def test_black(self):
+        im = pyvips.Image.black(100, 100)
+
+        assert im.width == 100
+        assert im.height == 100
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert im.bands == 1
+        for i in range(0, 100):
+            pixel = im(i, i)
+            assert len(pixel) == 1
+            assert pixel[0] == 0
+
+        im = pyvips.Image.black(100, 100, bands=3)
+
+        assert im.width == 100
+        assert im.height == 100
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert im.bands == 3
+        for i in range(0, 100):
+            pixel = im(i, i)
+            assert len(pixel) == 3
+            assert_almost_equal_objects(pixel, [0, 0, 0])
+
+    def test_buildlut(self):
+        M = pyvips.Image.new_from_array([[0, 0],
+                                         [255, 100]])
+        lut = M.buildlut()
+        assert lut.width == 256
+        assert lut.height == 1
+        assert lut.bands == 1
+        p = lut(0, 0)
+        assert p[0] == 0.0
+        p = lut(255, 0)
+        assert p[0] == 100.0
+        p = lut(10, 0)
+        assert p[0] == 100 * 10.0 / 255.0
+
+        M = pyvips.Image.new_from_array([[0, 0, 100],
+                                         [255, 100, 0],
+                                         [128, 10, 90]])
+        lut = M.buildlut()
+        assert lut.width == 256
+        assert lut.height == 1
+        assert lut.bands == 2
+        p = lut(0, 0)
+        assert_almost_equal_objects(p, [0.0, 100.0])
+        p = lut(64, 0)
+        assert_almost_equal_objects(p, [5.0, 95.0])
+
+    def test_eye(self):
+        im = pyvips.Image.eye(100, 90)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert im.max() == 1.0
+        assert im.min() == -1.0
+
+        im = pyvips.Image.eye(100, 90, uchar=True)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert im.max() == 255.0
+        assert im.min() == 0.0
+
+    def test_fractsurf(self):
+        im = pyvips.Image.fractsurf(100, 90, 2.5)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+    def test_gaussmat(self):
+        im = pyvips.Image.gaussmat(1, 0.1)
+        assert im.width == 5
+        assert im.height == 5
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.DOUBLE
+        assert im.max() == 20
+        total = im.avg() * im.width * im.height
+        scale = im.get("scale")
+        assert total == scale
+        p = im(im.width / 2, im.height / 2)
+        assert p[0] == 20.0
+
+        im = pyvips.Image.gaussmat(1, 0.1,
+                                   separable=True, precision="float")
+        assert im.width == 5
+        assert im.height == 1
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.DOUBLE
+        assert im.max() == 1.0
+        total = im.avg() * im.width * im.height
+        scale = im.get("scale")
+        assert total == scale
+        p = im(im.width / 2, im.height / 2)
+        assert p[0] == 1.0
+
+    def test_gaussnoise(self):
+        im = pyvips.Image.gaussnoise(100, 90)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+        im = pyvips.Image.gaussnoise(100, 90, sigma=10, mean=100)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+        sigma = im.deviate()
+        mean = im.avg()
+
+        assert pytest.approx(sigma, 0.2) == 10
+        assert pytest.approx(mean, 0.2) == 100
+
+    def test_grey(self):
+        im = pyvips.Image.grey(100, 90)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+        p = im(0, 0)
+        assert p[0] == 0.0
+        p = im(99, 0)
+        assert p[0] == 1.0
+        p = im(0, 89)
+        assert p[0] == 0.0
+        p = im(99, 89)
+        assert p[0] == 1.0
+
+        im = pyvips.Image.grey(100, 90, uchar=True)
+        assert im.width == 100
+        assert im.height == 90
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+
+        p = im(0, 0)
+        assert p[0] == 0
+        p = im(99, 0)
+        assert p[0] == 255
+        p = im(0, 89)
+        assert p[0] == 0
+        p = im(99, 89)
+        assert p[0] == 255
+
+    def test_identity(self):
+        im = pyvips.Image.identity()
+        assert im.width == 256
+        assert im.height == 1
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+
+        p = im(0, 0)
+        assert p[0] == 0.0
+        p = im(255, 0)
+        assert p[0] == 255.0
+        p = im(128, 0)
+        assert p[0] == 128.0
+
+        im = pyvips.Image.identity(ushort=True)
+        assert im.width == 65536
+        assert im.height == 1
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.USHORT
+
+        p = im(0, 0)
+        assert p[0] == 0
+        p = im(99, 0)
+        assert p[0] == 99
+        p = im(65535, 0)
+        assert p[0] == 65535
+
+    def test_invertlut(self):
+        lut = pyvips.Image.new_from_array([[0.1, 0.2, 0.3, 0.1],
+                                           [0.2, 0.4, 0.4, 0.2],
+                                           [0.7, 0.5, 0.6, 0.3]])
+        im = lut.invertlut()
+        assert im.width == 256
+        assert im.height == 1
+        assert im.bands == 3
+        assert im.format == pyvips.BandFormat.DOUBLE
+
+        p = im(0, 0)
+        assert_almost_equal_objects(p, [0, 0, 0])
+        p = im(255, 0)
+        assert_almost_equal_objects(p, [1, 1, 1])
+        p = im(0.2 * 255, 0)
+        assert pytest.approx(p[0], 0.1) == 0.1
+        p = im(0.3 * 255, 0)
+        assert pytest.approx(p[1], 0.1) == 0.1
+        p = im(0.1 * 255, 0)
+        assert pytest.approx(p[2], 0.1) == 0.1
+
+    def test_logmat(self):
+        im = pyvips.Image.logmat(1, 0.1)
+        assert im.width == 7
+        assert im.height == 7
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.DOUBLE
+        assert im.max() == 20
+        total = im.avg() * im.width * im.height
+        scale = im.get("scale")
+        assert total == scale
+        p = im(im.width / 2, im.height / 2)
+        assert p[0] == 20.0
+
+        im = pyvips.Image.logmat(1, 0.1,
+                                 separable=True, precision="float")
+        assert im.width == 7
+        assert im.height == 1
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.DOUBLE
+        assert im.max() == 1.0
+        total = im.avg() * im.width * im.height
+        scale = im.get("scale")
+        assert total == scale
+        p = im(im.width / 2, im.height / 2)
+        assert p[0] == 1.0
+
+    def test_mask_butterworth_band(self):
+        im = pyvips.Image.mask_butterworth_band(128, 128, 2,
+                                                0.5, 0.5, 0.7,
+                                                0.1)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert pytest.approx(im.max(), 0.01) == 1
+        p = im(32, 32)
+        assert p[0] == 1.0
+
+        im = pyvips.Image.mask_butterworth_band(128, 128, 2,
+                                                0.5, 0.5, 0.7,
+                                                0.1, uchar=True, optical=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert im.max() == 255
+        p = im(32, 32)
+        assert p[0] == 255.0
+        p = im(64, 64)
+        assert p[0] == 255.0
+
+        im = pyvips.Image.mask_butterworth_band(128, 128, 2,
+                                                0.5, 0.5, 0.7,
+                                                0.1, uchar=True, optical=True,
+                                                nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert im.max() == 255
+        p = im(32, 32)
+        assert p[0] == 255.0
+        p = im(64, 64)
+        assert p[0] != 255
+
+    def test_mask_butterworth(self):
+        im = pyvips.Image.mask_butterworth(128, 128, 2, 0.7, 0.1,
+                                           nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert pytest.approx(im.min(), 0.01) == 0
+        p = im(0, 0)
+        assert p[0] == 0.0
+        v, x, y = im.maxpos()
+        assert x == 64
+        assert y == 64
+
+        im = pyvips.Image.mask_butterworth(128, 128, 2, 0.7, 0.1,
+                                           optical=True, uchar=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert pytest.approx(im.min(), 0.01) == 0
+        p = im(64, 64)
+        assert p[0] == 255
+
+    def test_mask_butterworth_ring(self):
+        im = pyvips.Image.mask_butterworth_ring(128, 128, 2, 0.7, 0.1, 0.5,
+                                                nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        p = im(45, 0)
+        assert pytest.approx(p[0], 0.0001) == 1.0
+        v, x, y = im.minpos()
+        assert x == 64
+        assert y == 64
+
+    def test_mask_fractal(self):
+        im = pyvips.Image.mask_fractal(128, 128, 2.3)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+    def test_mask_gaussian_band(self):
+        im = pyvips.Image.mask_gaussian_band(128, 128, 0.5, 0.5, 0.7, 0.1)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert pytest.approx(im.max(), 0.01) == 1
+        p = im(32, 32)
+        assert p[0] == 1.0
+
+    def test_mask_gaussian(self):
+        im = pyvips.Image.mask_gaussian(128, 128, 0.7, 0.1,
+                                        nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert pytest.approx(im.min(), 0.01) == 0
+        p = im(0, 0)
+        assert p[0] == 0.0
+
+    def test_mask_gaussian_ring(self):
+        im = pyvips.Image.mask_gaussian_ring(128, 128, 0.7, 0.1, 0.5,
+                                             nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        p = im(45, 0)
+        assert pytest.approx(p[0], 0.001) == 1.0
+
+    def test_mask_ideal_band(self):
+        im = pyvips.Image.mask_ideal_band(128, 128, 0.5, 0.5, 0.7)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert pytest.approx(im.max(), 0.01) == 1
+        p = im(32, 32)
+        assert p[0] == 1.0
+
+    def test_mask_ideal(self):
+        im = pyvips.Image.mask_ideal(128, 128, 0.7,
+                                     nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        assert pytest.approx(im.min(), 0.01) == 0
+        p = im(0, 0)
+        assert p[0] == 0.0
+
+    def test_mask_gaussian_ring_2(self):
+        im = pyvips.Image.mask_ideal_ring(128, 128, 0.7, 0.5,
+                                          nodc=True)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+        p = im(45, 0)
+        assert pytest.approx(p[0], 0.001) == 1.0
+
+    def test_sines(self):
+        im = pyvips.Image.sines(128, 128)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+    @pytest.mark.skipif(pyvips.type_find("VipsOperation", "text") == 0,
+                        reason="no text, skipping test")
+    def test_text(self):
+        im = pyvips.Image.text("Hello, world!")
+        assert im.width > 10
+        assert im.height > 10
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.UCHAR
+        assert im.max() == 255
+        assert im.min() == 0
+
+    def test_tonelut(self):
+        im = pyvips.Image.tonelut()
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.USHORT
+        assert im.width == 32768
+        assert im.height == 1
+        assert im.hist_ismonotonic()
+
+    def test_xyz(self):
+        im = pyvips.Image.xyz(128, 128)
+        assert im.bands == 2
+        assert im.format == pyvips.BandFormat.UINT
+        assert im.width == 128
+        assert im.height == 128
+        p = im(45, 35)
+        assert_almost_equal_objects(p, [45, 35])
+
+    def test_zone(self):
+        im = pyvips.Image.zone(128, 128)
+        assert im.width == 128
+        assert im.height == 128
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+    @pytest.mark.skipif(pyvips.type_find("VipsOperation", "worley") == 0,
+                        reason="no worley, skipping test")
+    def test_worley(self):
+        im = pyvips.Image.worley(512, 512)
+        assert im.width == 512
+        assert im.height == 512
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+    @pytest.mark.skipif(pyvips.type_find("VipsOperation", "perlin") == 0,
+                        reason="no perlin, skipping test")
+    def test_perlin(self):
+        im = pyvips.Image.perlin(512, 512)
+        assert im.width == 512
+        assert im.height == 512
+        assert im.bands == 1
+        assert im.format == pyvips.BandFormat.FLOAT
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_draw.py
+++ b/test/test-suite/test_draw.py
@ -0,0 +1,101 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+
+
+class TestDraw:
+    def test_draw_circle(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(100, 50, 50, 25)
+        pixel = im(25, 50)
+        assert len(pixel) == 1
+        assert pixel[0] == 100
+        pixel = im(26, 50)
+        assert len(pixel) == 1
+        assert pixel[0] == 0
+
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(100, 50, 50, 25, fill=True)
+        pixel = im(25, 50)
+        assert len(pixel) == 1
+        assert pixel[0] == 100
+        pixel = im(26, 50)
+        assert pixel[0] == 100
+        pixel = im(24, 50)
+        assert pixel[0] == 0
+
+    def test_draw_flood(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(100, 50, 50, 25)
+        im = im.draw_flood(100, 50, 50)
+
+        im2 = pyvips.Image.black(100, 100)
+        im2 = im2.draw_circle(100, 50, 50, 25, fill=True)
+
+        diff = (im - im2).abs().max()
+        assert diff == 0
+
+    def test_draw_image(self):
+        im = pyvips.Image.black(51, 51)
+        im = im.draw_circle(100, 25, 25, 25, fill=True)
+
+        im2 = pyvips.Image.black(100, 100)
+        im2 = im2.draw_image(im, 25, 25)
+
+        im3 = pyvips.Image.black(100, 100)
+        im3 = im3.draw_circle(100, 50, 50, 25, fill=True)
+
+        diff = (im2 - im3).abs().max()
+        assert diff == 0
+
+    def test_draw_line(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_line(100, 0, 0, 100, 0)
+        pixel = im(0, 0)
+        assert len(pixel) == 1
+        assert pixel[0] == 100
+        pixel = im(0, 1)
+        assert len(pixel) == 1
+        assert pixel[0] == 0
+
+    def test_draw_mask(self):
+        mask = pyvips.Image.black(51, 51)
+        mask = mask.draw_circle(128, 25, 25, 25, fill=True)
+
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_mask(200, mask, 25, 25)
+
+        im2 = pyvips.Image.black(100, 100)
+        im2 = im2.draw_circle(100, 50, 50, 25, fill=True)
+
+        diff = (im - im2).abs().max()
+        assert diff == 0
+
+    def test_draw_rect(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_rect(100, 25, 25, 50, 50, fill=True)
+
+        im2 = pyvips.Image.black(100, 100)
+        for y in range(25, 75):
+            im2 = im2.draw_line(100, 25, y, 74, y)
+
+        diff = (im - im2).abs().max()
+        assert diff == 0
+
+    def test_draw_smudge(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(100, 50, 50, 25, fill=True)
+
+        im2 = im.draw_smudge(10, 10, 50, 50)
+
+        im3 = im.crop(10, 10, 50, 50)
+
+        im4 = im2.draw_image(im3, 10, 10)
+
+        diff = (im4 - im).abs().max()
+        assert diff == 0
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_foreign.py
+++ b/test/test-suite/test_foreign.py
@ -0,0 +1,787 @@
+# vim: set fileencoding=utf-8 :
+import gc
+import os
+import shutil
+import tempfile
+import pytest
+
+import pyvips
+from helpers import JPEG_FILE, SRGB_FILE, \
+    MATLAB_FILE, PNG_FILE, TIF_FILE, OME_FILE, ANALYZE_FILE, \
+    GIF_FILE, WEBP_FILE, EXR_FILE, FITS_FILE, OPENSLIDE_FILE, \
+    PDF_FILE, SVG_FILE, SVGZ_FILE, SVG_GZ_FILE, GIF_ANIM_FILE, \
+    DICOM_FILE, BMP_FILE, temp_filename, assert_almost_equal_objects, have, \
+    skip_if_no
+
+
+class TestForeign:
+    tempdir = None
+
+    @classmethod
+    def setup_class(cls):
+        cls.tempdir = tempfile.mkdtemp()
+
+        cls.colour = pyvips.Image.jpegload(JPEG_FILE)
+        cls.mono = cls.colour.extract_band(1)
+        # we remove the ICC profile: the RGB one will no longer be appropriate
+        cls.mono.remove("icc-profile-data")
+        cls.rad = cls.colour.float2rad()
+        cls.rad.remove("icc-profile-data")
+        cls.cmyk = cls.colour.bandjoin(cls.mono)
+        cls.cmyk = cls.cmyk.copy(interpretation=pyvips.Interpretation.CMYK)
+        cls.cmyk.remove("icc-profile-data")
+
+        im = pyvips.Image.new_from_file(GIF_FILE)
+        cls.onebit = im > 128
+
+    @classmethod
+    def teardown_class(cls):
+        shutil.rmtree(cls.tempdir, ignore_errors=True)
+
+    # we have test files for formats which have a clear standard
+    def file_loader(self, loader, test_file, validate):
+        im = pyvips.Operation.call(loader, test_file)
+        validate(im)
+        im = pyvips.Image.new_from_file(test_file)
+        validate(im)
+
+    def buffer_loader(self, loader, test_file, validate):
+        with open(test_file, 'rb') as f:
+            buf = f.read()
+
+        im = pyvips.Operation.call(loader, buf)
+        validate(im)
+        im = pyvips.Image.new_from_buffer(buf, "")
+        validate(im)
+
+    def save_load(self, format, im):
+        x = pyvips.Image.new_temp_file(format)
+        im.write(x)
+
+        assert im.width == x.width
+        assert im.height == x.height
+        assert im.bands == x.bands
+        max_diff = (im - x).abs().max()
+        assert max_diff == 0
+
+    def save_load_file(self, format, options, im, thresh):
+        # yuk!
+        # but we can't set format parameters for pyvips.Image.new_temp_file()
+        filename = temp_filename(self.tempdir, format)
+
+        im.write_to_file(filename + options)
+        x = pyvips.Image.new_from_file(filename)
+
+        assert im.width == x.width
+        assert im.height == x.height
+        assert im.bands == x.bands
+        max_diff = (im - x).abs().max()
+        assert max_diff <= thresh
+        x = None
+
+    def save_load_buffer(self, saver, loader, im, max_diff=0, **kwargs):
+        buf = pyvips.Operation.call(saver, im, **kwargs)
+        x = pyvips.Operation.call(loader, buf)
+
+        assert im.width == x.width
+        assert im.height == x.height
+        assert im.bands == x.bands
+        assert (im - x).abs().max() <= max_diff
+
+    def save_buffer_tempfile(self, saver, suf, im, max_diff=0):
+        filename = temp_filename(self.tempdir, suf)
+
+        buf = pyvips.Operation.call(saver, im)
+        f = open(filename, 'wb')
+        f.write(buf)
+        f.close()
+
+        x = pyvips.Image.new_from_file(filename)
+
+        assert im.width == x.width
+        assert im.height == x.height
+        assert im.bands == x.bands
+        assert (im - x).abs().max() <= max_diff
+
+    def test_vips(self):
+        self.save_load_file(".v", "", self.colour, 0)
+
+        # check we can save and restore metadata
+        filename = temp_filename(self.tempdir, ".v")
+        self.colour.write_to_file(filename)
+        x = pyvips.Image.new_from_file(filename)
+        before_exif = self.colour.get("exif-data")
+        after_exif = x.get("exif-data")
+
+        assert len(before_exif) == len(after_exif)
+        for i in range(len(before_exif)):
+            assert before_exif[i] == after_exif[i]
+
+        x = None
+
+    @skip_if_no("jpegload")
+    def test_jpeg(self):
+        def jpeg_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [6, 5, 3])
+            profile = im.get("icc-profile-data")
+            assert len(profile) == 1352
+            assert im.width == 1024
+            assert im.height == 768
+            assert im.bands == 3
+
+        self.file_loader("jpegload", JPEG_FILE, jpeg_valid)
+        self.save_load("%s.jpg", self.mono)
+        self.save_load("%s.jpg", self.colour)
+
+        self.buffer_loader("jpegload_buffer", JPEG_FILE, jpeg_valid)
+        self.save_load_buffer("jpegsave_buffer", "jpegload_buffer",
+                              self.colour, 80)
+
+        # see if we have exif parsing: our test image has this field
+        x = pyvips.Image.new_from_file(JPEG_FILE)
+        if x.get_typeof("exif-ifd0-Orientation") != 0:
+            # we need a copy of the image to set the new metadata on
+            # otherwise we get caching problems
+
+            # can set, save and load new orientation
+            x = pyvips.Image.new_from_file(JPEG_FILE)
+            x = x.copy()
+
+            x.set("orientation", 2)
+
+            filename = temp_filename(self.tempdir, '.jpg')
+            x.write_to_file(filename)
+
+            x = pyvips.Image.new_from_file(filename)
+            y = x.get("orientation")
+            assert y == 2
+
+            # can remove orientation, save, load again, orientation
+            # has reset
+            x.remove("orientation")
+
+            filename = temp_filename(self.tempdir, '.jpg')
+            x.write_to_file(filename)
+
+            x = pyvips.Image.new_from_file(filename)
+            y = x.get("orientation")
+            assert y == 1
+
+            # autorotate load works
+            x = pyvips.Image.new_from_file(JPEG_FILE)
+            x = x.copy()
+
+            x.set("orientation", 6)
+
+            filename = temp_filename(self.tempdir, '.jpg')
+            x.write_to_file(filename)
+
+            x1 = pyvips.Image.new_from_file(filename)
+            x2 = pyvips.Image.new_from_file(filename, autorotate=True)
+            assert x1.width == x2.height
+            assert x1.height == x2.width
+
+            # can set, save and reload ASCII string fields
+            x = pyvips.Image.new_from_file(JPEG_FILE)
+            x = x.copy()
+
+            x.set_type(pyvips.GValue.gstr_type, 
+                       "exif-ifd0-ImageDescription", "hello world")
+
+            filename = temp_filename(self.tempdir, '.jpg')
+            x.write_to_file(filename)
+
+            x = pyvips.Image.new_from_file(filename)
+            y = x.get("exif-ifd0-ImageDescription")
+            # can't use == since the string will have an extra " (xx, yy, zz)" 
+            # format area at the end
+            assert y.startswith("hello world")
+
+            # can set, save and reload UTF16 string fields ... pyvips is 
+            # utf8, but it will be coded as utf16 and back for the XP* fields
+            x = pyvips.Image.new_from_file(JPEG_FILE)
+            x = x.copy()
+
+            x.set_type(pyvips.GValue.gstr_type, "exif-ifd0-XPComment", "йцук")
+
+            filename = temp_filename(self.tempdir, '.jpg')
+            x.write_to_file(filename)
+
+            x = pyvips.Image.new_from_file(filename)
+            y = x.get("exif-ifd0-XPComment")
+            # can't use == since the string will have an extra " (xx, yy, zz)" 
+            # format area at the end
+            assert y.startswith("йцук")
+
+            # can set/save/load UserComment, a tag which has the
+            # encoding in the first 8 bytes ... though libexif only supports
+            # ASCII for this
+            x = pyvips.Image.new_from_file(JPEG_FILE)
+            x = x.copy()
+
+            x.set_type(pyvips.GValue.gstr_type, 
+                       "exif-ifd2-UserComment", "hello world")
+
+            filename = temp_filename(self.tempdir, '.jpg')
+            x.write_to_file(filename)
+
+            x = pyvips.Image.new_from_file(filename)
+            y = x.get("exif-ifd2-UserComment")
+            # can't use == since the string will have an extra " (xx, yy, zz)" 
+            # format area at the end
+            assert y.startswith("hello world")
+
+
+    @skip_if_no("pngload")
+    def test_png(self):
+        def png_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [38671.0, 33914.0, 26762.0])
+            assert im.width == 290
+            assert im.height == 442
+            assert im.bands == 3
+
+        self.file_loader("pngload", PNG_FILE, png_valid)
+        self.buffer_loader("pngload_buffer", PNG_FILE, png_valid)
+        self.save_load_buffer("pngsave_buffer", "pngload_buffer", self.colour)
+        self.save_load("%s.png", self.mono)
+        self.save_load("%s.png", self.colour)
+
+    @skip_if_no("tiffload")
+    def test_tiff(self):
+        def tiff_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [38671.0, 33914.0, 26762.0])
+            assert im.width == 290
+            assert im.height == 442
+            assert im.bands == 3
+
+        self.file_loader("tiffload", TIF_FILE, tiff_valid)
+        self.buffer_loader("tiffload_buffer", TIF_FILE, tiff_valid)
+        if pyvips.at_least_libvips(8, 5):
+            self.save_load_buffer("tiffsave_buffer",
+                                  "tiffload_buffer",
+                                  self.colour)
+        self.save_load("%s.tif", self.mono)
+        self.save_load("%s.tif", self.colour)
+        self.save_load("%s.tif", self.cmyk)
+
+        self.save_load("%s.tif", self.onebit)
+        self.save_load_file(".tif", "[squash]", self.onebit, 0)
+        self.save_load_file(".tif", "[miniswhite]", self.onebit, 0)
+        self.save_load_file(".tif", "[squash,miniswhite]", self.onebit, 0)
+
+        self.save_load_file(".tif",
+                            "[profile={0}]".format(SRGB_FILE),
+                            self.colour, 0)
+        self.save_load_file(".tif", "[tile]", self.colour, 0)
+        self.save_load_file(".tif", "[tile,pyramid]", self.colour, 0)
+        self.save_load_file(".tif",
+                            "[tile,pyramid,compression=jpeg]", self.colour, 80)
+        self.save_load_file(".tif", "[bigtiff]", self.colour, 0)
+        self.save_load_file(".tif", "[compression=jpeg]", self.colour, 80)
+        self.save_load_file(".tif",
+                            "[tile,tile-width=256]", self.colour, 10)
+
+        filename = temp_filename(self.tempdir, '.tif')
+        x = pyvips.Image.new_from_file(TIF_FILE)
+        x = x.copy()
+        x.set("orientation", 2)
+        x.write_to_file(filename)
+        x = pyvips.Image.new_from_file(filename)
+        y = x.get("orientation")
+        assert y == 2
+
+        filename = temp_filename(self.tempdir, '.tif')
+        x = pyvips.Image.new_from_file(TIF_FILE)
+        x = x.copy()
+        x.set("orientation", 2)
+        x.write_to_file(filename)
+        x = pyvips.Image.new_from_file(filename)
+        y = x.get("orientation")
+        assert y == 2
+        x.remove("orientation")
+
+        filename = temp_filename(self.tempdir, '.tif')
+        x.write_to_file(filename)
+        x = pyvips.Image.new_from_file(filename)
+        y = x.get("orientation")
+        assert y == 1
+
+        filename = temp_filename(self.tempdir, '.tif')
+        x = pyvips.Image.new_from_file(TIF_FILE)
+        x = x.copy()
+        x.set("orientation", 6)
+        x.write_to_file(filename)
+        x1 = pyvips.Image.new_from_file(filename)
+        x2 = pyvips.Image.new_from_file(filename, autorotate=True)
+        assert x1.width == x2.height
+        assert x1.height == x2.width
+
+        # OME support in 8.5
+        x = pyvips.Image.new_from_file(OME_FILE)
+        assert x.width == 439
+        assert x.height == 167
+        page_height = x.height
+
+        x = pyvips.Image.new_from_file(OME_FILE, n=-1)
+        assert x.width == 439
+        assert x.height == page_height * 15
+
+        x = pyvips.Image.new_from_file(OME_FILE, page=1, n=-1)
+        assert x.width == 439
+        assert x.height == page_height * 14
+
+        x = pyvips.Image.new_from_file(OME_FILE, page=1, n=2)
+        assert x.width == 439
+        assert x.height == page_height * 2
+
+        x = pyvips.Image.new_from_file(OME_FILE, n=-1)
+        assert x(0, 166)[0] == 96
+        assert x(0, 167)[0] == 0
+        assert x(0, 168)[0] == 1
+
+        filename = temp_filename(self.tempdir, '.tif')
+        x.write_to_file(filename)
+
+        x = pyvips.Image.new_from_file(filename, n=-1)
+        assert x.width == 439
+        assert x.height == page_height * 15
+        assert x(0, 166)[0] == 96
+        assert x(0, 167)[0] == 0
+        assert x(0, 168)[0] == 1
+
+        # pyr save to buffer added in 8.6
+        x = pyvips.Image.new_from_file(TIF_FILE)
+        buf = x.tiffsave_buffer(tile=True, pyramid=True)
+        filename = temp_filename(self.tempdir, '.tif')
+        x.tiffsave(filename, tile=True, pyramid=True)
+        with open(filename, 'rb') as f:
+            buf2 = f.read()
+        assert len(buf) == len(buf2)
+
+        a = pyvips.Image.new_from_buffer(buf, "", page=2)
+        b = pyvips.Image.new_from_buffer(buf2, "", page=2)
+        assert a.width == b.width
+        assert a.height == b.height
+        assert a.avg() == b.avg()
+
+        # region-shrink added in 8.7
+        x = pyvips.Image.new_from_file(TIF_FILE)
+        buf = x.tiffsave_buffer(tile=True, pyramid=True, region_shrink="mean")
+        buf = x.tiffsave_buffer(tile=True, pyramid=True, region_shrink="mode")
+        buf = x.tiffsave_buffer(tile=True, pyramid=True, region_shrink="median")
+
+    @skip_if_no("magickload")
+    def test_magickload(self):
+        def bmp_valid(im):
+            a = im(100, 100)
+
+            assert_almost_equal_objects(a, [227, 216, 201])
+            assert im.width == 1419
+            assert im.height == 1001
+
+        self.file_loader("magickload", BMP_FILE, bmp_valid)
+        self.buffer_loader("magickload_buffer", BMP_FILE, bmp_valid)
+
+        # we should have rgba for svg files
+        im = pyvips.Image.magickload(SVG_FILE)
+        assert im.bands == 4
+
+        # density should change size of generated svg
+        im = pyvips.Image.magickload(SVG_FILE, density='100')
+        width = im.width
+        height = im.height
+        im = pyvips.Image.magickload(SVG_FILE, density='200')
+        # This seems to fail on travis, no idea why, some problem in their IM
+        # perhaps
+        # assert im.width == width * 2
+        # assert im.height == height * 2
+
+        # all-frames should load every frame of the animation
+        # (though all-frames is deprecated)
+        im = pyvips.Image.magickload(GIF_ANIM_FILE)
+        width = im.width
+        height = im.height
+        im = pyvips.Image.magickload(GIF_ANIM_FILE, all_frames=True)
+        assert im.width == width
+        assert im.height == height * 5
+
+        # page/n let you pick a range of pages
+        # 'n' param added in 8.5
+        if pyvips.at_least_libvips(8, 5):
+            im = pyvips.Image.magickload(GIF_ANIM_FILE)
+            width = im.width
+            height = im.height
+            im = pyvips.Image.magickload(GIF_ANIM_FILE, page=1, n=2)
+            assert im.width == width
+            assert im.height == height * 2
+            page_height = im.get("page-height")
+            assert page_height == height
+
+        # should work for dicom
+        im = pyvips.Image.magickload(DICOM_FILE)
+        assert im.width == 128
+        assert im.height == 128
+        # some IMs are 3 bands, some are 1, can't really test
+        # assert im.bands == 1
+
+        # added in 8.7
+        if have("magicksave"):
+            self.save_load_file(".bmp", "", self.colour, 0)
+            self.save_load_buffer("magicksave_buffer", "magickload_buffer",
+                                  self.colour, 0, format="BMP")
+            self.save_load("%s.bmp", self.colour)
+
+    @skip_if_no("webpload")
+    def test_webp(self):
+        def webp_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [71, 166, 236])
+            assert im.width == 550
+            assert im.height == 368
+            assert im.bands == 3
+
+        self.file_loader("webpload", WEBP_FILE, webp_valid)
+        self.buffer_loader("webpload_buffer", WEBP_FILE, webp_valid)
+        self.save_load_buffer("webpsave_buffer", "webpload_buffer",
+                              self.colour, 60)
+        self.save_load("%s.webp", self.colour)
+
+        # test lossless mode
+        im = pyvips.Image.new_from_file(WEBP_FILE)
+        buf = im.webpsave_buffer(lossless=True)
+        im2 = pyvips.Image.new_from_buffer(buf, "")
+        assert im.avg() == im2.avg()
+
+        # higher Q should mean a bigger buffer
+        b1 = im.webpsave_buffer(Q=10)
+        b2 = im.webpsave_buffer(Q=90)
+        assert len(b2) > len(b1)
+
+        # try saving an image with an ICC profile and reading it back ... if we
+        # can do it, our webp supports metadata load/save
+        buf = self.colour.webpsave_buffer()
+        im = pyvips.Image.new_from_buffer(buf, "")
+        if im.get_typeof("icc-profile-data") != 0:
+            # verify that the profile comes back unharmed
+            p1 = self.colour.get("icc-profile-data")
+            p2 = im.get("icc-profile-data")
+            assert p1 == p2
+
+            # add tests for exif, xmp, ipct
+            # the exif test will need us to be able to walk the header,
+            # we can't just check exif-data
+
+            # we can test that exif changes change the output of webpsave
+            # first make sure we have exif support
+            z = pyvips.Image.new_from_file(JPEG_FILE)
+            if z.get_typeof("exif-ifd0-Orientation") != 0:
+                x = self.colour.copy()
+                x.set("orientation", 6)
+                buf = x.webpsave_buffer()
+                y = pyvips.Image.new_from_buffer(buf, "")
+                assert y.get("orientation") == 6
+
+    @skip_if_no("analyzeload")
+    def test_analyzeload(self):
+        def analyze_valid(im):
+            a = im(10, 10)
+            assert pytest.approx(a[0]) == 3335
+            assert im.width == 128
+            assert im.height == 8064
+            assert im.bands == 1
+
+        self.file_loader("analyzeload", ANALYZE_FILE, analyze_valid)
+
+    @skip_if_no("matload")
+    def test_matload(self):
+        def matlab_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [38671.0, 33914.0, 26762.0])
+            assert im.width == 290
+            assert im.height == 442
+            assert im.bands == 3
+
+        self.file_loader("matload", MATLAB_FILE, matlab_valid)
+
+    @skip_if_no("openexrload")
+    def test_openexrload(self):
+        def exr_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [0.124512, 0.159668,
+                                            0.040375, 1.0],
+                                        threshold=0.00001)
+            assert im.width == 610
+            assert im.height == 406
+            assert im.bands == 4
+
+        self.file_loader("openexrload", EXR_FILE, exr_valid)
+
+    @skip_if_no("fitsload")
+    def test_fitsload(self):
+        def fits_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [-0.165013, -0.148553, 1.09122,
+                                            -0.942242], threshold=0.00001)
+            assert im.width == 200
+            assert im.height == 200
+            assert im.bands == 4
+
+        self.file_loader("fitsload", FITS_FILE, fits_valid)
+        self.save_load("%s.fits", self.mono)
+
+    @skip_if_no("openslideload")
+    def test_openslideload(self):
+        def openslide_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [244, 250, 243, 255])
+            assert im.width == 2220
+            assert im.height == 2967
+            assert im.bands == 4
+
+        self.file_loader("openslideload", OPENSLIDE_FILE, openslide_valid)
+
+    @skip_if_no("pdfload")
+    def test_pdfload(self):
+        def pdf_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [35, 31, 32, 255])
+            assert im.width == 1133
+            assert im.height == 680
+            assert im.bands == 4
+
+        self.file_loader("pdfload", PDF_FILE, pdf_valid)
+        self.buffer_loader("pdfload_buffer", PDF_FILE, pdf_valid)
+
+        im = pyvips.Image.new_from_file(PDF_FILE)
+        x = pyvips.Image.new_from_file(PDF_FILE, scale=2)
+        assert abs(im.width * 2 - x.width) < 2
+        assert abs(im.height * 2 - x.height) < 2
+
+        im = pyvips.Image.new_from_file(PDF_FILE)
+        x = pyvips.Image.new_from_file(PDF_FILE, dpi=144)
+        assert abs(im.width * 2 - x.width) < 2
+        assert abs(im.height * 2 - x.height) < 2
+
+    @skip_if_no("gifload")
+    def test_gifload(self):
+        def gif_valid(im):
+            a = im(10, 10)
+            assert_almost_equal_objects(a, [33])
+            assert im.width == 159
+            assert im.height == 203
+            assert im.bands == 1
+
+        self.file_loader("gifload", GIF_FILE, gif_valid)
+        self.buffer_loader("gifload_buffer", GIF_FILE, gif_valid)
+
+        # 'n' param added in 8.5
+        if pyvips.at_least_libvips(8, 5):
+            x1 = pyvips.Image.new_from_file(GIF_ANIM_FILE)
+            x2 = pyvips.Image.new_from_file(GIF_ANIM_FILE, n=2)
+            assert x2.height == 2 * x1.height
+            page_height = x2.get("page-height")
+            assert page_height == x1.height
+
+            x2 = pyvips.Image.new_from_file(GIF_ANIM_FILE, n=-1)
+            assert x2.height == 5 * x1.height
+
+            x2 = pyvips.Image.new_from_file(GIF_ANIM_FILE, page=1, n=-1)
+            assert x2.height == 4 * x1.height
+
+    @skip_if_no("svgload")
+    def test_svgload(self):
+        def svg_valid(im):
+            a = im(10, 10)
+
+            # some old rsvg versions are way, way off
+            assert abs(a[0] - 79) < 2
+            assert abs(a[1] - 79) < 2
+            assert abs(a[2] - 132) < 2
+            assert abs(a[3] - 255) < 2
+
+            assert im.width == 288
+            assert im.height == 470
+            assert im.bands == 4
+
+        self.file_loader("svgload", SVG_FILE, svg_valid)
+        self.buffer_loader("svgload_buffer", SVG_FILE, svg_valid)
+
+        self.file_loader("svgload", SVGZ_FILE, svg_valid)
+        self.buffer_loader("svgload_buffer", SVGZ_FILE, svg_valid)
+
+        self.file_loader("svgload", SVG_GZ_FILE, svg_valid)
+
+        im = pyvips.Image.new_from_file(SVG_FILE)
+        x = pyvips.Image.new_from_file(SVG_FILE, scale=2)
+        assert abs(im.width * 2 - x.width) < 2
+        assert abs(im.height * 2 - x.height) < 2
+
+        im = pyvips.Image.new_from_file(SVG_FILE)
+        x = pyvips.Image.new_from_file(SVG_FILE, dpi=144)
+        assert abs(im.width * 2 - x.width) < 2
+        assert abs(im.height * 2 - x.height) < 2
+
+    def test_csv(self):
+        self.save_load("%s.csv", self.mono)
+
+    def test_matrix(self):
+        self.save_load("%s.mat", self.mono)
+
+    @skip_if_no("ppmload")
+    def test_ppm(self):
+        self.save_load("%s.ppm", self.mono)
+        self.save_load("%s.ppm", self.colour)
+
+    @skip_if_no("radload")
+    def test_rad(self):
+        self.save_load("%s.hdr", self.colour)
+        self.save_buffer_tempfile("radsave_buffer", ".hdr",
+                                  self.rad, max_diff=0)
+
+    @skip_if_no("dzsave")
+    def test_dzsave(self):
+        # dzsave is hard to test, there are so many options
+        # test each option separately and hope they all function together
+        # correctly
+
+        # default deepzoom layout ... we must use png here, since we want to
+        # test the overlap for equality
+        filename = temp_filename(self.tempdir, '')
+        self.colour.dzsave(filename, suffix=".png")
+
+        # test horizontal overlap ... expect 256 step, overlap 1
+        x = pyvips.Image.new_from_file(filename + "_files/10/0_0.png")
+        assert x.width == 255
+        y = pyvips.Image.new_from_file(filename + "_files/10/1_0.png")
+        assert y.width == 256
+
+        # the right two columns of x should equal the left two columns of y
+        left = x.crop(x.width - 2, 0, 2, x.height)
+        right = y.crop(0, 0, 2, y.height)
+        assert (left - right).abs().max() == 0
+
+        # test vertical overlap
+        assert x.height == 255
+        y = pyvips.Image.new_from_file(filename + "_files/10/0_1.png")
+        assert y.height == 256
+
+        # the bottom two rows of x should equal the top two rows of y
+        top = x.crop(0, x.height - 2, x.width, 2)
+        bottom = y.crop(0, 0, y.width, 2)
+        assert (top - bottom).abs().max() == 0
+
+        # there should be a bottom layer
+        x = pyvips.Image.new_from_file(filename + "_files/0/0_0.png")
+        assert x.width == 1
+        assert x.height == 1
+
+        # 10 should be the final layer
+        assert not os.path.isdir(filename + "_files/11")
+
+        # default google layout
+        filename = temp_filename(self.tempdir, '')
+        self.colour.dzsave(filename, layout="google")
+
+        # test bottom-right tile ... default is 256x256 tiles, overlap 0
+        x = pyvips.Image.new_from_file(filename + "/2/2/3.jpg")
+        assert x.width == 256
+        assert x.height == 256
+        assert not os.path.exists(filename + "/2/2/4.jpg")
+        assert not os.path.exists(filename + "/3")
+        x = pyvips.Image.new_from_file(filename + "/blank.png")
+        assert x.width == 256
+        assert x.height == 256
+
+        # google layout with overlap ... verify that we clip correctly
+
+        # overlap 1, 510x510 pixels, 256 pixel tiles, should be exactly 2x2
+        # tiles, though in fact the bottom and right edges will be white
+        filename = temp_filename(self.tempdir, '')
+        self.colour.crop(0, 0, 510, 510).dzsave(filename, layout="google",
+                                                overlap=1, depth="one")
+
+        x = pyvips.Image.new_from_file(filename + "/0/1/1.jpg")
+        assert x.width == 256
+        assert x.height == 256
+        assert not os.path.exists(filename + "/0/2/2.jpg")
+
+        # with 511x511, it'll fit exactly into 2x2 -- we we actually generate
+        # 3x3, since we output the overlaps
+        filename = temp_filename(self.tempdir, '')
+        self.colour.crop(0, 0, 511, 511).dzsave(filename, layout="google",
+                                                overlap=1, depth="one")
+
+        x = pyvips.Image.new_from_file(filename + "/0/2/2.jpg")
+        assert x.width == 256
+        assert x.height == 256
+        assert not os.path.exists(filename + "/0/3/3.jpg")
+
+        # default zoomify layout
+        filename = temp_filename(self.tempdir, '')
+        self.colour.dzsave(filename, layout="zoomify")
+
+        # 256x256 tiles, no overlap
+        assert os.path.exists(filename + "/ImageProperties.xml")
+        x = pyvips.Image.new_from_file(filename + "/TileGroup0/2-3-2.jpg")
+        assert x.width == 256
+        assert x.height == 256
+
+        # test zip output
+        filename = temp_filename(self.tempdir, '.zip')
+        self.colour.dzsave(filename)
+        assert os.path.exists(filename)
+        assert not os.path.exists(filename + "_files")
+        assert not os.path.exists(filename + ".dzi")
+
+        # test compressed zip output
+        filename2 = temp_filename(self.tempdir, '.zip')
+        self.colour.dzsave(filename2, compression=-1)
+        assert os.path.exists(filename2)
+        assert os.path.getsize(filename2) < os.path.getsize(filename)
+
+        # test suffix
+        filename = temp_filename(self.tempdir, '')
+        self.colour.dzsave(filename, suffix=".png")
+
+        x = pyvips.Image.new_from_file(filename + "_files/10/0_0.png")
+        assert x.width == 255
+
+        # test overlap
+        filename = temp_filename(self.tempdir, '')
+        self.colour.dzsave(filename, overlap=200)
+
+        y = pyvips.Image.new_from_file(filename + "_files/10/1_1.jpeg")
+        assert y.width == 654
+
+        # test tile-size
+        filename = temp_filename(self.tempdir, '')
+        self.colour.dzsave(filename, tile_size=512)
+
+        y = pyvips.Image.new_from_file(filename + "_files/10/0_0.jpeg")
+        assert y.width == 513
+        assert y.height == 513
+
+        # test save to memory buffer
+        filename = temp_filename(self.tempdir, '.zip')
+        base = os.path.basename(filename)
+        root, ext = os.path.splitext(base)
+
+        self.colour.dzsave(filename)
+        with open(filename, 'rb') as f:
+            buf1 = f.read()
+        buf2 = self.colour.dzsave_buffer(basename=root)
+        assert len(buf1) == len(buf2)
+
+        # we can't test the bytes are exactly equal -- the timestamps will
+        # be different
+
+        # added in 8.7
+        buf = self.colour.dzsave_buffer(region_shrink="mean")
+        buf = self.colour.dzsave_buffer(region_shrink="mode")
+        buf = self.colour.dzsave_buffer(region_shrink="median")
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_histogram.py
+++ b/test/test-suite/test_histogram.py
@ -0,0 +1,114 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+from helpers import JPEG_FILE
+
+
+class TestHistogram:
+    def test_hist_cum(self):
+        im = pyvips.Image.identity()
+
+        sum = im.avg() * 256
+
+        cum = im.hist_cum()
+
+        p = cum(255, 0)
+        assert p[0] == sum
+
+    def test_hist_equal(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+
+        im2 = im.hist_equal()
+
+        assert im.width == im2.width
+        assert im.height == im2.height
+
+        assert im.avg() < im2.avg()
+        assert im.deviate() < im2.deviate()
+
+    def test_hist_ismonotonic(self):
+        im = pyvips.Image.identity()
+        assert im.hist_ismonotonic()
+
+    def test_hist_local(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+
+        im2 = im.hist_local(10, 10)
+
+        assert im.width == im2.width
+        assert im.height == im2.height
+
+        assert im.avg() < im2.avg()
+        assert im.deviate() < im2.deviate()
+
+        if pyvips.at_least_libvips(8, 5):
+            im3 = im.hist_local(10, 10, max_slope=3)
+
+            assert im.width == im3.width
+            assert im.height == im3.height
+
+            assert im3.deviate() < im2.deviate()
+
+    def test_hist_match(self):
+        im = pyvips.Image.identity()
+        im2 = pyvips.Image.identity()
+
+        matched = im.hist_match(im2)
+
+        assert (im - matched).abs().max() == 0.0
+
+    def test_hist_norm(self):
+        im = pyvips.Image.identity()
+        im2 = im.hist_norm()
+
+        assert (im - im2).abs().max() == 0.0
+
+    def test_hist_plot(self):
+        im = pyvips.Image.identity()
+        im2 = im.hist_plot()
+
+        assert im2.width == 256
+        assert im2.height == 256
+        assert im2.format == pyvips.BandFormat.UCHAR
+        assert im2.bands == 1
+
+    def test_hist_map(self):
+        im = pyvips.Image.identity()
+
+        im2 = im.maplut(im)
+
+        assert (im - im2).abs().max() == 0.0
+
+    def test_percent(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE).extract_band(1)
+
+        pc = im.percent(90)
+
+        msk = im <= pc
+        n_set = (msk.avg() * msk.width * msk.height) / 255.0
+        pc_set = 100 * n_set / (msk.width * msk.height)
+
+        assert pytest.approx(pc_set, 0.5) == 90
+
+    def test_hist_entropy(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE).extract_band(1)
+
+        ent = im.hist_find().hist_entropy()
+
+        assert pytest.approx(ent, 0.01) == 4.37
+
+    def test_stdif(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+
+        im2 = im.stdif(10, 10)
+
+        assert im.width == im2.width
+        assert im.height == im2.height
+
+        # new mean should be closer to target mean
+        assert abs(im.avg() - 128) > abs(im2.avg() - 128)
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_iofuncs.py
+++ b/test/test-suite/test_iofuncs.py
@ -0,0 +1,93 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+from helpers import assert_equal_objects
+
+
+class TestIofuncs:
+    # test the vips7 filename splitter ... this is very fragile and annoying
+    # code with lots of cases
+    def test_split7(self):
+        def split(path):
+            filename7 = pyvips.path_filename7(path)
+            mode7 = pyvips.path_mode7(path)
+
+            return [filename7, mode7]
+
+        cases = [
+            ["c:\\silly:dir:name\\fr:ed.tif:jpeg:95,,,,c:\\icc\\srgb.icc",
+             ["c:\\silly:dir:name\\fr:ed.tif",
+              "jpeg:95,,,,c:\\icc\\srgb.icc"]],
+            ["I180:",
+             ["I180",
+              ""]],
+            ["c:\\silly:",
+             ["c:\\silly",
+              ""]],
+            ["c:\\program files\\x:hello",
+             ["c:\\program files\\x",
+              "hello"]],
+            ["C:\\fixtures\\2569067123_aca715a2ee_o.jpg",
+             ["C:\\fixtures\\2569067123_aca715a2ee_o.jpg",
+              ""]]
+        ]
+
+        for case in cases:
+            assert_equal_objects(split(case[0]), case[1])
+
+    def test_new_from_image(self):
+        im = pyvips.Image.mask_ideal(100, 100, 0.5,
+                                     reject=True, optical=True)
+
+        im2 = im.new_from_image(12)
+
+        assert im2.width == im.width
+        assert im2.height == im.height
+        assert im2.interpretation == im.interpretation
+        assert im2.format == im.format
+        assert im2.xres == im.xres
+        assert im2.yres == im.yres
+        assert im2.xoffset == im.xoffset
+        assert im2.yoffset == im.yoffset
+        assert im2.bands == 1
+        assert im2.avg() == 12
+
+        im2 = im.new_from_image([1, 2, 3])
+
+        assert im2.bands == 3
+        assert im2.avg() == 2
+
+    def test_new_from_memory(self):
+        s = bytearray(200)
+        im = pyvips.Image.new_from_memory(s, 20, 10, 1, 'uchar')
+
+        assert im.width == 20
+        assert im.height == 10
+        assert im.format == 'uchar'
+        assert im.bands == 1
+        assert im.avg() == 0
+
+        im += 10
+
+        assert im.avg() == 10
+
+    @pytest.mark.skipif(not pyvips.at_least_libvips(8, 5),
+                        reason="requires libvips >= 8.5")
+    def test_get_fields(self):
+        im = pyvips.Image.black(10, 10)
+        fields = im.get_fields()
+        # we might add more fields later
+        assert len(fields) > 10
+        assert fields[0] == 'width'
+
+    def test_write_to_memory(self):
+        s = bytearray(200)
+        im = pyvips.Image.new_from_memory(s, 20, 10, 1, 'uchar')
+        t = im.write_to_memory()
+
+        assert s == t
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_morphology.py
+++ b/test/test-suite/test_morphology.py
@ -0,0 +1,55 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+
+
+class TestMorphology:
+    def test_countlines(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_line(255, 0, 50, 100, 50)
+        n_lines = im.countlines(pyvips.Direction.HORIZONTAL)
+        assert n_lines == 1
+
+    def test_labelregions(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(255, 50, 50, 25, fill=True)
+        mask, opts = im.labelregions(segments=True)
+
+        assert opts['segments'] == 3
+        assert mask.max() == 2
+
+    def test_erode(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(255, 50, 50, 25, fill=True)
+        im2 = im.erode([[128, 255, 128],
+                        [255, 255, 255],
+                        [128, 255, 128]])
+        assert im.width == im2.width
+        assert im.height == im2.height
+        assert im.bands == im2.bands
+        assert im.avg() > im2.avg()
+
+    def test_dilate(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(255, 50, 50, 25, fill=True)
+        im2 = im.dilate([[128, 255, 128],
+                         [255, 255, 255],
+                         [128, 255, 128]])
+        assert im.width == im2.width
+        assert im.height == im2.height
+        assert im.bands == im2.bands
+        assert im2.avg() > im.avg()
+
+    def test_rank(self):
+        im = pyvips.Image.black(100, 100)
+        im = im.draw_circle(255, 50, 50, 25, fill=True)
+        im2 = im.rank(3, 3, 8)
+        assert im.width == im2.width
+        assert im.height == im2.height
+        assert im.bands == im2.bands
+        assert im2.avg() > im.avg()
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/test-suite/test_resample.py
+++ b/test/test-suite/test_resample.py
@ -0,0 +1,203 @@
+# vim: set fileencoding=utf-8 :
+import pytest
+
+import pyvips
+from helpers import JPEG_FILE, all_formats, have
+
+
+# Run a function expecting a complex image on a two-band image
+def run_cmplx(fn, image):
+    if image.format == pyvips.BandFormat.FLOAT:
+        new_format = pyvips.BandFormat.COMPLEX
+    elif image.format == pyvips.BandFormat.DOUBLE:
+        new_format = pyvips.BandFormat.DPCOMPLEX
+    else:
+        raise pyvips.Error("run_cmplx: not float or double")
+
+    # tag as complex, run, revert tagging
+    cmplx = image.copy(bands=1, format=new_format)
+    cmplx_result = fn(cmplx)
+
+    return cmplx_result.copy(bands=2, format=image.format)
+
+
+def to_polar(image):
+    """Transform image coordinates to polar.
+
+    The image is transformed so that it is wrapped around a point in the
+    centre. Vertical straight lines become circles or segments of circles,
+    horizontal straight lines become radial spokes.
+    """
+    # xy image, zero in the centre, scaled to fit image to a circle
+    xy = pyvips.Image.xyz(image.width, image.height)
+    xy -= [image.width / 2.0, image.height / 2.0]
+    scale = min(image.width, image.height) / float(image.width)
+    xy *= 2.0 / scale
+
+    # to polar, scale vertical axis to 360 degrees
+    index = run_cmplx(lambda x: x.polar(), xy)
+    index *= [1, image.height / 360.0]
+
+    return image.mapim(index)
+
+
+def to_rectangular(image):
+    """Transform image coordinates to rectangular.
+
+    The image is transformed so that it is unwrapped from a point in the
+    centre. Circles or segments of circles become vertical straight lines,
+    radial lines become horizontal lines.
+    """
+    # xy image, vertical scaled to 360 degrees
+    xy = pyvips.Image.xyz(image.width, image.height)
+    xy *= [1, 360.0 / image.height]
+
+    # to rect, scale to image rect
+    index = run_cmplx(lambda x: x.rect(), xy)
+    scale = min(image.width, image.height) / float(image.width)
+    index *= scale / 2.0
+    index += [image.width / 2.0, image.height / 2.0]
+
+    return image.mapim(index)
+
+
+class TestResample:
+    def test_affine(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+
+        # vsqbs is non-interpolatory, don't test this way
+        for name in ["nearest", "bicubic", "bilinear", "nohalo", "lbb"]:
+            x = im
+            interpolate = pyvips.Interpolate.new(name)
+            for i in range(4):
+                x = x.affine([0, 1, 1, 0], interpolate=interpolate)
+
+            assert (x - im).abs().max() == 0
+
+    def test_reduce(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+        # cast down to 0-127, the smallest range, so we aren't messed up by
+        # clipping
+        im = im.cast(pyvips.BandFormat.CHAR)
+
+        for fac in [1, 1.1, 1.5, 1.999]:
+            for fmt in all_formats:
+                for kernel in ["nearest", "linear",
+                               "cubic", "lanczos2", "lanczos3"]:
+                    x = im.cast(fmt)
+                    r = x.reduce(fac, fac, kernel=kernel)
+                    d = abs(r.avg() - im.avg())
+                    assert d < 2
+
+        # try constant images ... should not change the constant
+        for const in [0, 1, 2, 254, 255]:
+            im = (pyvips.Image.black(10, 10) + const).cast("uchar")
+            for kernel in ["nearest", "linear",
+                           "cubic", "lanczos2", "lanczos3"]:
+                # print "testing kernel =", kernel
+                # print "testing const =", const
+                shr = im.reduce(2, 2, kernel=kernel)
+                d = abs(shr.avg() - im.avg())
+                assert d == 0
+
+    def test_resize(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+        im2 = im.resize(0.25)
+        assert im2.width == round(im.width / 4.0)
+        assert im2.height == round(im.height / 4.0)
+
+        # test geometry rounding corner case
+        im = pyvips.Image.black(100, 1)
+        x = im.resize(0.5)
+        assert x.width == 50
+        assert x.height == 1
+
+    def test_shrink(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+        im2 = im.shrink(4, 4)
+        assert im2.width == round(im.width / 4.0)
+        assert im2.height == round(im.height / 4.0)
+        assert abs(im.avg() - im2.avg()) < 1
+
+        im2 = im.shrink(2.5, 2.5)
+        assert im2.width == round(im.width / 2.5)
+        assert im2.height == round(im.height / 2.5)
+        assert abs(im.avg() - im2.avg()) < 1
+
+    @pytest.mark.skipif(not pyvips.at_least_libvips(8, 5),
+                        reason="requires libvips >= 8.5")
+    def test_thumbnail(self):
+        im = pyvips.Image.thumbnail(JPEG_FILE, 100)
+
+        assert im.width == 100
+        assert im.bands == 3
+        assert im.bands == 3
+
+        # the average shouldn't move too much
+        im_orig = pyvips.Image.new_from_file(JPEG_FILE)
+        assert abs(im_orig.avg() - im.avg()) < 1
+
+        # make sure we always get the right width
+        for width in range(1000, 1, -13):
+            im = pyvips.Image.thumbnail(JPEG_FILE, width)
+            assert im.width == width
+
+        # should fit one of width or height
+        im = pyvips.Image.thumbnail(JPEG_FILE, 100, height=300)
+        assert im.width == 100
+        assert im.height != 300
+        im = pyvips.Image.thumbnail(JPEG_FILE, 300, height=100)
+        assert im.width != 300
+        assert im.height == 100
+
+        # with @crop, should fit both width and height
+        im = pyvips.Image.thumbnail(JPEG_FILE, 100,
+                                    height=300, crop=True)
+        assert im.width == 100
+        assert im.height == 300
+
+        im1 = pyvips.Image.thumbnail(JPEG_FILE, 100)
+        with open(JPEG_FILE, 'rb') as f:
+            buf = f.read()
+        im2 = pyvips.Image.thumbnail_buffer(buf, 100)
+        assert abs(im1.avg() - im2.avg()) < 1
+
+    def test_similarity(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+        im2 = im.similarity(angle=90)
+        im3 = im.affine([0, -1, 1, 0])
+        # rounding in calculating the affine transform from the angle stops
+        # this being exactly true
+        assert (im2 - im3).abs().max() < 50
+
+    def test_similarity_scale(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+        im2 = im.similarity(scale=2)
+        im3 = im.affine([2, 0, 0, 2])
+        assert (im2 - im3).abs().max() == 0
+
+    # added in 8.7
+    def test_rotate(self):
+        if have("rotate"):
+            im = pyvips.Image.new_from_file(JPEG_FILE)
+            im2 = im.rotate(90)
+            im3 = im.affine([0, -1, 1, 0])
+            # rounding in calculating the affine transform from the angle stops
+            # this being exactly true
+            assert (im2 - im3).abs().max() < 50
+
+    def test_mapim(self):
+        im = pyvips.Image.new_from_file(JPEG_FILE)
+
+        p = to_polar(im)
+        r = to_rectangular(p)
+
+        # the left edge (which is squashed to the origin) will be badly
+        # distorted, but the rest should not be too bad
+        a = r.crop(50, 0, im.width - 50, im.height).gaussblur(2)
+        b = im.crop(50, 0, im.width - 50, im.height).gaussblur(2)
+        assert (a - b).abs().max() < 20
+
+
+if __name__ == '__main__':
+    pytest.main()
--- a/test/variables.sh.in
+++ b/test/variables.sh.in
@ -2,7 +2,7 @@ top_srcdir=@TOP_SRCDIR@
 PYTHON=@PYTHON@
 # we need a different tmp for each script since make can run tests in parallel
 tmp=$top_srcdir/test/tmp-$$
-test_images=$top_srcdir/test/images
+test_images=$top_srcdir/test/test-suite/images
 image=$test_images/йцук.jpg
 mkdir -p $tmp
 vips=$top_srcdir/tools/vips