2013-08-07 08:44:08 +02:00
|
|
|
<?php
|
2013-08-07 22:59:09 +02:00
|
|
|
$pomo = dirname( dirname( dirname( __FILE__ ) ) ) . '/src/wp-includes/pomo';
|
2013-08-07 08:44:08 +02:00
|
|
|
require_once "$pomo/entry.php";
|
|
|
|
require_once "$pomo/translations.php";
|
|
|
|
|
2013-10-26 02:41:34 +02:00
|
|
|
/**
|
|
|
|
* Responsible for extracting translatable strings from PHP source files
|
|
|
|
* in the form of Translations instances
|
|
|
|
*/
|
2013-08-07 08:44:08 +02:00
|
|
|
class StringExtractor {
|
|
|
|
|
2013-10-26 02:45:16 +02:00
|
|
|
var $rules = array(
|
|
|
|
'__' => array( 'string' ),
|
|
|
|
'_e' => array( 'string' ),
|
|
|
|
'_n' => array( 'singular', 'plural' ),
|
|
|
|
);
|
2013-08-07 08:44:08 +02:00
|
|
|
var $comment_prefix = 'translators:';
|
|
|
|
|
|
|
|
function __construct( $rules = array() ) {
|
|
|
|
$this->rules = $rules;
|
|
|
|
}
|
|
|
|
|
|
|
|
function extract_from_directory( $dir, $excludes = array(), $includes = array(), $prefix = '' ) {
|
|
|
|
$old_cwd = getcwd();
|
|
|
|
chdir( $dir );
|
|
|
|
$translations = new Translations;
|
|
|
|
$file_names = (array) scandir( '.' );
|
|
|
|
foreach ( $file_names as $file_name ) {
|
|
|
|
if ( '.' == $file_name || '..' == $file_name ) continue;
|
|
|
|
if ( preg_match( '/\.php$/', $file_name ) && $this->does_file_name_match( $prefix . $file_name, $excludes, $includes ) ) {
|
|
|
|
$translations->merge_originals_with( $this->extract_from_file( $file_name, $prefix ) );
|
|
|
|
}
|
|
|
|
if ( is_dir( $file_name ) ) {
|
|
|
|
$translations->merge_originals_with( $this->extract_from_directory( $file_name, $excludes, $includes, $prefix . $file_name . '/' ) );
|
|
|
|
}
|
|
|
|
}
|
|
|
|
chdir( $old_cwd );
|
|
|
|
return $translations;
|
|
|
|
}
|
|
|
|
|
|
|
|
function extract_from_file( $file_name, $prefix ) {
|
|
|
|
$code = file_get_contents( $file_name );
|
2013-10-26 02:54:27 +02:00
|
|
|
return $this->extract_from_code( $code, $prefix . $file_name );
|
2013-08-07 08:44:08 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
function does_file_name_match( $path, $excludes, $includes ) {
|
|
|
|
if ( $includes ) {
|
|
|
|
$matched_any_include = false;
|
|
|
|
foreach( $includes as $include ) {
|
|
|
|
if ( preg_match( '|^'.$include.'$|', $path ) ) {
|
|
|
|
$matched_any_include = true;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if ( !$matched_any_include ) return false;
|
|
|
|
}
|
|
|
|
if ( $excludes ) {
|
|
|
|
foreach( $excludes as $exclude ) {
|
|
|
|
if ( preg_match( '|^'.$exclude.'$|', $path ) ) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
function entry_from_call( $call, $file_name ) {
|
|
|
|
$rule = isset( $this->rules[$call['name']] )? $this->rules[$call['name']] : null;
|
|
|
|
if ( !$rule ) return null;
|
|
|
|
$entry = new Translation_Entry;
|
|
|
|
$multiple = array();
|
|
|
|
$complete = false;
|
|
|
|
for( $i = 0; $i < count( $rule ); ++$i ) {
|
|
|
|
if ( $rule[$i] && ( !isset( $call['args'][$i] ) || !is_string( $call['args'][$i] ) || '' == $call['args'][$i] ) ) return false;
|
|
|
|
switch( $rule[$i] ) {
|
|
|
|
case 'string':
|
|
|
|
if ( $complete ) {
|
|
|
|
$multiple[] = $entry;
|
|
|
|
$entry = new Translation_Entry;
|
|
|
|
$complete = false;
|
|
|
|
}
|
|
|
|
$entry->singular = $call['args'][$i];
|
|
|
|
$complete = true;
|
|
|
|
break;
|
|
|
|
case 'singular':
|
|
|
|
if ( $complete ) {
|
|
|
|
$multiple[] = $entry;
|
|
|
|
$entry = new Translation_Entry;
|
|
|
|
$complete = false;
|
|
|
|
}
|
|
|
|
$entry->singular = $call['args'][$i];
|
|
|
|
$entry->is_plural = true;
|
|
|
|
break;
|
|
|
|
case 'plural':
|
|
|
|
$entry->plural = $call['args'][$i];
|
|
|
|
$entry->is_plural = true;
|
|
|
|
$complete = true;
|
|
|
|
break;
|
|
|
|
case 'context':
|
|
|
|
$entry->context = $call['args'][$i];
|
|
|
|
foreach( $multiple as &$single_entry ) {
|
|
|
|
$single_entry->context = $entry->context;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if ( isset( $call['line'] ) && $call['line'] ) {
|
|
|
|
$references = array( $file_name . ':' . $call['line'] );
|
|
|
|
$entry->references = $references;
|
|
|
|
foreach( $multiple as &$single_entry ) {
|
|
|
|
$single_entry->references = $references;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if ( isset( $call['comment'] ) && $call['comment'] ) {
|
|
|
|
$comments = rtrim( $call['comment'] ) . "\n";
|
|
|
|
$entry->extracted_comments = $comments;
|
|
|
|
foreach( $multiple as &$single_entry ) {
|
|
|
|
$single_entry->extracted_comments = $comments;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if ( $multiple && $entry ) {
|
|
|
|
$multiple[] = $entry;
|
|
|
|
return $multiple;
|
|
|
|
}
|
|
|
|
|
|
|
|
return $entry;
|
|
|
|
}
|
|
|
|
|
2013-10-26 02:54:27 +02:00
|
|
|
function extract_from_code( $code, $file_name ) {
|
2013-08-07 08:44:08 +02:00
|
|
|
$translations = new Translations;
|
|
|
|
$function_calls = $this->find_function_calls( array_keys( $this->rules ), $code );
|
|
|
|
foreach( $function_calls as $call ) {
|
|
|
|
$entry = $this->entry_from_call( $call, $file_name );
|
|
|
|
if ( is_array( $entry ) )
|
|
|
|
foreach( $entry as $single_entry )
|
|
|
|
$translations->add_entry_or_merge( $single_entry );
|
|
|
|
elseif ( $entry)
|
|
|
|
$translations->add_entry_or_merge( $entry );
|
|
|
|
}
|
|
|
|
return $translations;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Finds all function calls in $code and returns an array with an associative array for each function:
|
|
|
|
* - name - name of the function
|
|
|
|
* - args - array for the function arguments. Each string literal is represented by itself, other arguments are represented by null.
|
|
|
|
* - line - line number
|
|
|
|
*/
|
|
|
|
function find_function_calls( $function_names, $code ) {
|
|
|
|
$tokens = token_get_all( $code );
|
|
|
|
$function_calls = array();
|
|
|
|
$latest_comment = false;
|
|
|
|
$in_func = false;
|
|
|
|
foreach( $tokens as $token ) {
|
|
|
|
$id = $text = null;
|
|
|
|
if ( is_array( $token ) ) list( $id, $text, $line ) = $token;
|
|
|
|
if ( T_WHITESPACE == $id ) continue;
|
|
|
|
if ( T_STRING == $id && in_array( $text, $function_names ) && !$in_func ) {
|
|
|
|
$in_func = true;
|
|
|
|
$paren_level = -1;
|
|
|
|
$args = array();
|
|
|
|
$func_name = $text;
|
|
|
|
$func_line = $line;
|
|
|
|
$func_comment = $latest_comment? $latest_comment : '';
|
|
|
|
|
|
|
|
$just_got_into_func = true;
|
|
|
|
$latest_comment = false;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if ( T_COMMENT == $id ) {
|
|
|
|
$text = trim( preg_replace( '%^/\*|//%', '', preg_replace( '%\*/$%', '', $text ) ) );
|
|
|
|
if ( 0 === stripos( $text, $this->comment_prefix ) ) {
|
|
|
|
$latest_comment = $text;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if ( !$in_func ) continue;
|
|
|
|
if ( '(' == $token ) {
|
|
|
|
$paren_level++;
|
|
|
|
if ( 0 == $paren_level ) { // start of first argument
|
|
|
|
$just_got_into_func = false;
|
|
|
|
$current_argument = null;
|
|
|
|
$current_argument_is_just_literal = true;
|
|
|
|
}
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if ( $just_got_into_func ) {
|
|
|
|
// there wasn't a opening paren just after the function name -- this means it is not a function
|
|
|
|
$in_func = false;
|
|
|
|
$just_got_into_func = false;
|
|
|
|
}
|
|
|
|
if ( ')' == $token ) {
|
|
|
|
if ( 0 == $paren_level ) {
|
|
|
|
$in_func = false;
|
|
|
|
$args[] = $current_argument;
|
|
|
|
$call = array( 'name' => $func_name, 'args' => $args, 'line' => $func_line );
|
|
|
|
if ( $func_comment ) $call['comment'] = $func_comment;
|
|
|
|
$function_calls[] = $call;
|
|
|
|
}
|
|
|
|
$paren_level--;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if ( ',' == $token && 0 == $paren_level ) {
|
|
|
|
$args[] = $current_argument;
|
|
|
|
$current_argument = null;
|
|
|
|
$current_argument_is_just_literal = true;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if ( T_CONSTANT_ENCAPSED_STRING == $id && $current_argument_is_just_literal ) {
|
|
|
|
// we can use eval safely, because we are sure $text is just a string literal
|
|
|
|
eval('$current_argument = '.$text.';' );
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
$current_argument_is_just_literal = false;
|
|
|
|
$current_argument = null;
|
|
|
|
}
|
|
|
|
return $function_calls;
|
|
|
|
}
|
|
|
|
}
|