Problems related to WordPress PHPUnit tests

[roberthusak]

This issue summarizes all the problems which cause WordPress PHPUnit tests to end prematurely (uncaught .NET exception, stack overflow etc.) or pollute output with extra text. Currently, they are solved by workarounds - by commenting certain parts out or by skipping. Some of them might require further investigation.

• [x] function_exists( 'add_filter' ) (see here) is replaced by true even though add_filter (defined in PeachPied.WordPress) might not be defined in the current Context if it's definition hasn't been included yet.

• [ ] symlink function is defined but throws NotSupportedException, which is not caught by PHPUnit.

• [x] imagecopyresampled doesn't support negative $src_w and $src_h. They should flip the image, see here.

• [x] DOMDocument::loadHTML throws an exception when parsing / in <input readonly/> - it thinks that readonly/ is a (obviously invalid) name of a parameter.

• [x] XSLTProcessor::importStylesheet doesn't support simplexml.

• [ ] getid3_quicktime.QuicktimeParseAtom (see here) occupies 267 200 bytes on the call stack, causing stack overflow.

• [x] When $x is e.g. stdClass, $x[0] causes InvalidOperationException, unable to be caught by PHPUnit.

• [ ] It seems that the MySQL connections are not closed (maybe due to a single Context for all tests and different behaviour of destructors?), reaching default number of max_connections option in about the third of the tests (returning the error: Too many connections). The current workaround is to increase the option manually.

• [x] file_exists doesn't take into account source files contained in PHAR archives. (it should use internal StreamWrapper API, the same for SplFileInfo, stat, etc.)

• [ ] As wp-includes/class-json.php is excluded from WordPress compilation due to the missing PEAR_Error class, the test of Services_JSON must be skipped for now.

• [ ] Class GP_Locale is dynamically downloaded and included (see here), which currently causes its definition to be output. Including it in the tests statically should prevent this from happening.

[jakubmisek]

@roberthusak can we implement symlink using Mono.Unix.UnixSymbolicLinkInfo.CreateSymbolicLinkTo() ?

[jakubmisek]

@roberthusak I think $x[0] causes a fatal error in PHP which won't be handled either? Isn't it a different bug in PeachPie?

[roberthusak]

@roberthusak can we implement symlink using Mono.Unix.UnixSymbolicLinkInfo.CreateSymbolicLinkTo() ?

@jakubmisek That's true, we can make it work the same way as chown. If we want to make it run on Windows as well (PHP does that since Windows Vista), we'll just need to find an appropriate way (maybe using P/Invoke as here), as .NET Standard seems not to have the API yet.

@roberthusak I think $x[0] causes a fatal error in PHP which won't be handled either? Isn't it a different bug in PeachPie?

In fact, it throws a catchable Error (it seems the fatal error is caused by not catching it), try this code (tested on PHP 7.4.1):

function test($a) {
  try {
    echo $a[0];
  } catch (Throwable $e) {
    echo get_class($e);
  }
}

test(new stdClass());

The reason why this error occurred at all is probably a bug in Peachpie (or some kind of incompatibility), which will require further investigation (as well as in the case of other unsuccessful tests).

[jakubmisek]

Note - large getid3_quicktime.QuicktimeParseAtom stack size might be caused when debugging since the compiler emits IndirectLocal for each local variable to be displayed in IDE's "Locals" window.

This might help: https://github.com/peachpiecompiler/peachpie/commit/df2266aa88ef0f0ef0a235fc97e55a00df34c975

[roberthusak]

Note - large getid3_quicktime.QuicktimeParseAtom stack size might be caused when debugging since the compiler emits IndirectLocal for each local variable to be displayed in IDE's "Locals" window.

This might help: df2266a

Unfortunately, this doesn't help. The stack overflow still happens even with the latest commit of Peachpie and even in Release.

[jakubmisek]

issue with stack size in getid3_quicktime.QuicktimeParseAtom is related to https://github.com/dotnet/runtime/issues/8980

The method is too large for JIT. The solution would be for us to emit fewer temp value types on the stack.