- In the first one we were calculating the tmp_len wrong which made the
u_strFromUTF32() function try to convert too many code points.
- The second issue was a bit more subtle as the "what" string wasn't
duplicated but still modified. This string is passed as data to the
function and this kind of data the engine tries to free when the function
ends. Because we were re-allocating the data the original memory location
was already freed resulting in a double free error when the engine tries to
free the argument as it was passed to the function.
fgets() will work now as will anything which calls one of the
_php_stream_get_line() family of functions.
The one exception here is when the legacy defines are used on a unicode
stream. At the moment they'll simply return NULL, I'll update these
to do sloppy conversion in a bit.
'make (u)test' still doesn't work, but it's a different doesn't work.
<?php
declare(encoding="latin1");
$a = "1234å67890";
file_put_contents( "/tmp/testuc.1", $a);
file_put_contents( "/tmp/testuc.2", (string) $a);
$context = stream_context_create();
stream_context_set_params($context, array( "output_encoding" => "latin1" ) );
file_put_contents( "/tmp/testuc.3", $a, FILE_TEXT, $context);
file_put_contents( "/tmp/testuc.4", (string) $a, FILE_TEXT, $context);
?>
But it still throws a warning on ".3". It's a small design issue that I
didn't want to touch right now.
Don't be frightened by the size of this commit.
A significant portion of it is restoring the read buffer semantics back
to what PHP4/5 use. (Or a close aproximation thereof).
See main/streams/streams.c and ext/standard/file.c for a set of
UTODO comments covering work yet to be done.
