Short-lived regression from 5ce9687cb2.
I forgot to add the persistent local flag, so that means that RC_DEBUG
will complain. These strings are local to the thread so we can just add
the flag to silence the debug checker in this case.
The usage of the current API within an observer handler leads to bugs like https://bugs.xdebug.org/view.php?id=2232.
Given two observer handlers, next to each other. The first one is executed. It removes itself. The second observer handler gets moved to the place of the first. The first one returns. The handler in the second slot is fetched, but is now NULL, because the it's now in the slot of the first observer; i.e. the second handler is skipped and the begin/end symmetry guarantee is violated.
Providing the next handler to the caller is a zero-cost way to avoid any impact in the paths of zend_observe_fcall_begin/end.
Signed-off-by: Bob Weinand <bobwei9@hotmail.com>
When the attribute has a single text child, we can avoid an allocating
call to libxml2 and read the contents directly.
On my i7-4790, I tested the optimization with both the $value and
$nodeValue property.
```
Summary
./sapi/cli/php bench_value.php ran
1.82 ± 0.09 times faster than ./sapi/cli/php_old bench_value.php
Summary
./sapi/cli/php bench_nodeValue.php ran
1.78 ± 0.10 times faster than ./sapi/cli/php_old bench_nodeValue.php
```
Test code:
```
$dom = new DOMDocument;
$dom->loadXML('<root attrib="this is a relatively short text"/>');
$attrib = $dom->documentElement->attributes[0];
for ($i=0; $i<1000*1000; $i++) {
$attrib->value; // or nodeValue
}
```
For now, working on Linux, FreeBSD >= 13.x and DragonFlyBSD.
Handy wrapper to assign an array of cpu ids or to retrieve the cpu ids
assigned to a given process.
pcntl_setaffinity inserts valid unique cpu ids (within the range of available
cpus).
Close GH-13893
The implementation is needlessly obfuscated. It's not immediately clear that
MODMULT is a simple modular multiplication, despite its name. Specifically it's
not clear which of the parameters is the second factor.
Furthermore the stated period is off-by-one: A value of `0` is part of its own
chain, thus it may not be included in the period of the underlying generators.
The hash tables used are allocated via the persistent allocator.
When using ini_set, the allocation happens via the non-persistent
allocator. When the table is then freed in GSHUTDOWN, we get a crash
because the allocators are mismatched.
As a side note, it is strange that this is designed this way, because it
means that ini_sets persist between requests...
Co-authored-by: Kamil Tekiela <tekiela246@gmail.com>
The x86-32 build uses a fast path, but when I disabled the fast path I
got the following compile error with -Werror:
mysqlnd_portability.h:221:95: error: right shift count >= width of type [-Werror=shift-count-overflow]
For 32-bit platforms that don't have the fast path, this can cause
undefined behaviour at runtime. Some CPUs just mask the shift amount
so in those cases you even get wrong results.
This is not always found during the build because -Werror is off by
default.
`end = *p+maxlen`, and pointer overflow is UB, so that means that a check
of the form `end < *p` will always be false because it can only be true
on pointer overflow. In particular, the compiler simplifies this to
`maxlen < 0` which is always false because maxlen is unsigned.
When all costs are equal, levenshtein fulfills the requirements
of being a metric. A metric is symmetric, so we can swap the strings in
that case. Since we use rows of a partial matrix of length |string2| we
can make the choice of using string1 instead if |string1| < |string2|,
which will optimize memory usage and CPU time.
