If O_TNMPFILE is available, use it. This is a linux extension that
creates an unlinked file, so it'll be closed by the system when the last
associated fd is closed.
see https://github.com/jcupitt/libvips/pull/930
Before, they could make B_W for one-band output. This caused problems
with (for example) two black image bandjoined: the second band then
looked like an alpha to hasalpha() and enabled premultiply/unpremultiply
for operations like affine.
Now, it's always MULTIBAND. This is the generic multiband image type, so
you don't get any unexpected alpha handling.
The result of hasalpha is used to turn on things like
premultiplication, so we should be rather conservative
about when we signal this. We don't want to premultiply
things that should not be premultiplied.
Check Type as well as bands.
See: https://github.com/jcupitt/libvips/issues/918
libpng has started throwing hard errors if the profile does not match
the image -- this can happen all the time with perofiles inherited from
images that have been processed.
Test profiles before save and drop them (with a warning) if they are
incompatible with the image.
vips__get_bytes() used to fail if the file was too small for the
allocated buffer, which was a problem for svg, since files can be extremely
small.
This change makes vips__get_bytes() return the number of bytes read, so
the is_a testers can work on files smaller than the max header size.
icc_import can take a fallback input profile in case the embedded one is
broken or missing. If we use the fallback profile, this change attaches
it to the output image.
This means that icc_import will always output an image with the icc
profile that was used to import it. This helps to make the behaviour of
`thumbnail` more consistent.
See https://github.com/jcupitt/libvips/issues/152
If a delayed load failed, it could leave the pipeline only half-set up.
Sebsequent threads could then segv.
Set a load-has-failed flag and test before generate.
See https://github.com/jcupitt/libvips/issues/893
We were rounding up with ceil() when intize-ing convolution masks.
However, the vector path has a true range of (1.0, -1.0], so a mask with
1.0 as the max (for example) was actually triggering the overflow detector
and falling back to the C path.
Round up with ceil(x + 1) instead, so 1.0 (for example) will be mapped
to 0.5 and won't overflow.
