It should be incredibly rare for something to fail only without opcache while
also being platform specific. We're keeping opcache-less tests for x64 Linux,
macOS and i386 will continue to be tested without opcache in nightly.
Closes GH-11867
Commit 0b2e6bc2b0 started caching the directory entry type to improve
performance. Shortly after, we've seen flaky failures of the
buildFromIterator phar test.
When it fails, it's always a value error in the constructor of
RecursiveDirectoryIterator::__construct() with a "no such file or
directory" error. What's happening here is this:
1) A parallel test creates a subdirectory in the current working dir.
2) This test checks hasChildren() on a directory entry, the cached entry
returns "yes" on the subdirectory.
3) The parallel test finishes and removes the subdirectory.
4) The constructor mentioned above is called, causing an exception
because the directory is gone.
This race has always been possible, even before said commit. It's just
that it was very hard to hit before: the expensive stat call made the
race window hard to hit. The race is now easier to hit because of the
caching that is fast.
Since there's many tests that modify the current working directory, it
seems best to mark this as an "all" conflict. We cannot avoid every
TOC-TOU race when working with files with these phar tests.
In particular, mounteddir.phpt caused every conflict I saw on CI, but
there's more tests that create subdirectories in the current working
directory.
Closes GH-11869.
People relied on manually waiting for children, but the fix for GH-11498
broke this. Fixing this in PHP is fundamentally incompatible with doing
the wait loop in userland. This reverts to the old behaviour.
Closes GH-11863.
According to https://www.php.net/manual/en/class.domdocument:
When using json_encode() on a DOMDocument object the result will be
that of encoding an empty object.
But this was broken in 8.1. The output was `{"config": null}`.
That's because the config property is defined with a default value of
NULL, hence it was included. The other properties are not included
because they don't have a default property, and nothing is ever written
to their backing field. Hence, the JSON encoder excludes them.
Similarly, `(array) $doc` would yield the same `config` key in the
array.
Closes GH-11840.
It looks like the config.w32 uses CHECK_HEADER_ADD_INCLUDE to add the include
path to libxml into the search path.
That doesn't happen in zend-test.
To add to the Windows trouble, libxml is statically linked in, ext/libxml can
only be built statically but ext/zend-test can be built both statically and
dynamically.
So the regression tests won't work in all possible configurations anyway on Windows.
All of this is no problem on Linux because it just uses dynamic linking
and pkg-config, without any magic.
Signed-off-by: Ben Ramsey <ramsey@php.net>
Fixes GHSA-3qrf-m4j2-pcrr.
To parse a document with libxml2, you first need to create a parsing context.
The parsing context contains parsing options (e.g. XML_NOENT to substitute
entities) that the application (in this case PHP) can set.
Unfortunately, libxml2 also supports providing default set options.
For example, if you call xmlSubstituteEntitiesDefault(1) then the XML_NOENT
option will be added to the parsing options every time you create a parsing
context **even if the application never requested XML_NOENT**.
Third party extensions can override these globals, in particular the
substitute entity global. This causes entity substitution to be
unexpectedly active.
Fix it by setting the parsing options to a sane known value.
For API calls that depend on global state we introduce
PHP_LIBXML_SANITIZE_GLOBALS() and PHP_LIBXML_RESTORE_GLOBALS().
For other APIs that work directly with a context we introduce
php_libxml_sanitize_parse_ctxt_options().
At one point this was changed from a bool to an int in libxml2, with
negative values meaning it is unspecified. Because it is cast to a bool
this therefore returned true instead of the expected false.
Closes GH-11793.
- publicId could crash PHP if none was provided
- notationName never worked
The fields of this classs were untested. This new test file changes that.
Closes GH-11779.
There are a couple of oddities.
1) The interned strings buffer comprises the whole hashtable
datastructure.
Therefore, it seems that the interned strings buffer size is the size of
only said table. However, in the current code it also includes the size
of the zend_accel_shared_globals.
2) ZCSG(interned_strings).end is computed starting from the accelerator
globals struct itself. I would expect it to start from the part where
the interned strings table starts.
3) When computing the used size, it is done using
ZCSG(interned_strings).end - ZCSG(interned_strings).start. However,
this does not include the uin32_t slots array because
ZCSG(interned_strings).start pointers after that array.
This patch corrrects these 3 points.
Closes GH-11717.
Previously, FFI_G(symbols) and FFI_G(tags) were never cleaned up when calling
new on an existing object. However, if cdef() is called without parameters these
globals are NULL and might be created when new() creates new definitions. These
would then be discarded without freeing them.
Closes GH-11751
Not sure if this is possible to hit in practice, zend_accel_error_noreturn
doesn't return so the unlock isn't called. Other callsites that use both
zend_accel_error_noreturn and zend_shared_alloc_unlock first perform the
unlocking.
Closes GH-11718.
This also includes a fix for the MySQL ND driver to actually respect the user decided behaviour.
Closes GH-11622
Signed-off-by: George Peter Banyard <girgias@php.net>
Niels Dossche