The file_cache_only option causes the storage to be per process,
furthermore the arena is destroyed per request. Thus, zend_string's
can't survive between request and the permanent flag should not
be set. This is already done with the file cache part, but the
persistency part is used in various scenarios and should respect
this case as well. In this particular bug, the pcre pattern cache
needs to survive between requests and uses pattern strings as hash
keys. One more case relevant here would be various situations where
the flow disables the use of shared memory.
The fix is based on the same strategy for handling namespace
declarations as used by getAttributeNode. Note that this strategy makes
these methods not return a DOMAttr for xmlns* attributes, but an
instance of the (undocumented) class DOMNameSpaceNode. This is not
really ideal, but at least this fix makes the behavior of
getAttributeNode and getAttributeNodeNS consistent.
A follow-up action would be to investigate whether DOMNameSpaceNode can
be made into a subclass of DOMAttr (which may be hard due to the way
libxml treats namespace declarations) or document this deviating return
value for xmlns* attributes.
This bug had already been fixed, but apparently there's no regression
test yet, so we add one.
Note that the expected image has black pixel artifacts, which are
another issue (perhaps bug #40158), and would have to be adressed
separately.
We must take into account the line padding, when we're reading XBM
files.
We deliberately ignore the potential integer overflow here, because
that would be caught by gdImageCreate() or even earlier if `bytes==0`,
what happens in libgd00094.phpt which we adapt accordingly.
GD2 stores the number of horizontal and vertical chunks as words (i.e. 2
byte unsigned). These values are multiplied and assigned to an int when
reading the image, what can cause integer overflows. We have to avoid
that, and also make sure that either chunk count is actually greater
than zero. If illegal chunk counts are detected, we bail out from
reading the image.
(cherry picked from commit 5b5d9db3988b829e0b121b74bb3947f01c2796a1)
We must not pretend that there are image data if there are none. Instead
we fail reading the image file gracefully.
(cherry picked from commit cdb648dc4115ce0722f3cc75e6a65115fc0e56ab)
opcache_reset() only schedules the restart. Under circumstances,
the follow up requests might run uncached, until the restart
condition is met. To mitigate the false positives caused by this
behavior, any tests using opcache_reset() should not be put in
between other tests. Thus, moving the corresponding test to be
executed last.
Nikita Popov