We also add zend_map_ptr_static, so that we do not incur the overhead of constantly recreating the internal run_time_cache pointers on each request.
This mechanism might be extended for mutable_data of internal classes too.
Inline the lookup whether a function is observed at all.
This strategy is also used for FRAMELESS calls. If the frameless call is observed, we instead allocate a call frame and push the arguments, to call the the function afterwards.
Doing so is still a performance benefit as opposed to executing individual INIT_FCALL+SEND_VAL ops. Thus, even if the frameless call turns out to be observed, the call overhead is slightly lower than before.
If the internal function is not observed at all, the unavoidable overhead is fetching the FLF zend_function pointer and the run-time cache needs to be inspected.
As part of this work, it turned out to be most viable to put the result operand on the ZEND_OP_DATA instead of ZEND_FRAMELESS_ICALL_3, allowing seamless interoperability with the DO_ICALL opcode.
This is a bit unusual in comparison to all other ZEND_OP_DATA usages, but seems to not pose problems overall.
There is also a small issue resolved: trampolines would always use the ZEND_CALL_TRAMPOLINE_SPEC_OBSERVER function due to zend_observer_fcall_op_array_extension being set to -1 too late.
`zend_strtod.c` uses a global state (mostly an allocation freelist) protected by a mutex in ZTS builds. This state is used by `zend_dtoa()`, `zend_strtod()`, and variants. This creates a lot of contention in concurrent loads. `zend_dtoa()` is used to format floats to string, e.g. in sprintf, json_encode, serialize, uniqid.
Here I move the global state to the thread specific `executor_globals` and remove the mutex.
The impact on non-concurrent environments is null or negligible, but there is a considerable speed up on concurrent environments, especially on Alpine/Musl.
Struct members require some alignment based on their type. This means
that if a struct member is not aligned, there will be a hole created by
the compiler in the struct, which is wasted space. This patch reorders
some of the most commonly used structs, but in such a way that the
fields which were in the same cache line still belong together.
The only exception to this is exception_ignore_args, which was
temporally not close to nearby members, and as such I placed
it further up to close a hole.
On 64-bit Linux this gives us the following shrinks:
* zend_op_array: 248 -> 240
* zend_ssa_var: 56 -> 48
* zend_ssa_var_info: 48 -> 40
* php_core_globals: 672 -> 608
* zend_executor_globals: 1824 -> 1792
On 32-bit, the sizes will either remain the same or will result in
smaller shrinks.
`zend_uchar` suggests that the value is an ASCII character, but here,
it's about very small integers. This is misleading, so let's use a
C99 integer instead.
On all architectures currently supported by PHP, `zend_uchar` and
`uint8_t` are identical. This change is only about code readability.
`zend_long` is a signed integer that's only 32 bit on some 64 bit
architectures (e.g. ARM64). The proper type for memory sizes is
`size_t`, and this type is accepted by zend_fiber_init_context().
* Stop copying internal functions into each thread
It seems we don't copy internal methods for a long time, so this
shouldn't be a problem. We had to copy functions in PHP-5 times, but it
seems we just forgot to remove this.
It's possible that some third-part extensions (e.g. profilers, tracers,
debuggers) modify internal functions. After this change that may cause
race conditions in ZTS build (but we already jave the same behavior for
internal methods). Observer API should provide necesssary functionality
to avoid shared structures modification.
* Remove unused function
This reverts commit d0527427be.
This patch makes Swoole/Swow can not work anymore, because Coroutine will yield to another one during socket operation, EG(record_errors) assertion will always fail, and zend_begin_record_errors() was only used during compile time before.
Note: zend_emit_recorded_errors() and the typo fix are reserved.
This is not actually related to SSL handshake but stream socket creation
which does not clean errors if the error handler is set. This fix
prevents emitting errors until the stream is freed.
This is done by adding a new zend_atomic_bool type. The type
definition is only available for compiler alignment and size info; it
should be treated as opaque and only the zend_atomic_bool_* family of
functions should be used.
Note that directly using atomic_bool is complicated. All C++ compilers
stdlibs that I checked typedef atomic_bool to std::atomic<bool>, which
can't be used in an extern "C" section, and there's at least one usage
of this in core, and probably more outside of it.
So, instead use platform specific functions, preferring compiler
intrinsics.
If we only store the biased pointer, the map ptr region will not
be recognized as reachable memory by leak checkers. This is
primarily problematic for fuzzing, because this is persistent
memory that may be reallocated during the request, without being
an actual leak.
Avoid this by simply storing both the real base pointer of the
allocation, as well as the biased base pointer used for accesses.
This removes switching to main for fatal errors in fibers in favor of catching any zend_bailout in a fiber and calling zend_bailout again after switching to the previous fiber or {main}.
Since 3e6b447979 it is again possible to have
warnings (deprecations) during inheritance, and more such functionality is
likely in the future. This is a problem, because such warnings will only be
shown on the first request if the opcache inheritance cache is used. This
currently causes test failures in --repeat builds.
Fix this by uplifting the error recording functionality from opcache to Zend,
and then using it to persist a warning trace in the inheritance cache, which
can then be used to replay the warnings on subsequent executions.
This is needed by both fibers and opcache (and GH-6903 also uses it),
so make it a common structure that can be used by any functionality
storing warnings/errors.
This PR corrects misspellings identified by the check-spelling action.
The misspellings have been reported at jsoref@b6ba3e2#commitcomment-48946465
The action reports that the changes in this PR would make it happy: jsoref@602417c
Closes GH-6822.
Some upcoming changes like https://wiki.php.net/rfc/deprecate_null_to_scalar_internal_arg
will make it somewhat inconvenient to determine whether a given
function invocation will generate a diagnostic. Rather than trying
to exclude this in advance, call the function with diagnostics
suppressed, and check whether anything was thrown.
This adds a new EG flag that is kept specific to the SCCP use-case.
This does not use the error_cb hook as it is a (non-TLS) global,
and doesn't fully suppress error handling besides.
Test this by removing the in advance checks for implode and array_flip.
This is a new transparent technology that eliminates overhead of PHP class inheritance.
PHP classes are compiled and cached (by opcahce) separately, however their "linking" was done at run-time - on each request. The process of "linking" may involve a number of compatibility checks and borrowing methods/properties/constants form parent and traits. This takes significant time, but the result is the same on each request.
Inheritance Cache performs "linking" for unique set of all the depending classes (parent, interfaces, traits, property types, method types involved into compatibility checks) once and stores result in opcache shared memory. As a part of the this patch, I removed limitations for immutable classes (unresolved constants, typed properties and covariant type checks). So now all classes stored in opcache are "immutable". They may be lazily loaded into process memory, if necessary, but this usually occurs just once (on first linking).
The patch shows 8% improvement on Symphony "Hello World" app.
We're starting to see a mix between uses of zend_bool and bool.
Replace all usages with the standard bool type everywhere.
Of course, zend_bool is retained as an alias.
Add a `zend.exception_string_param_max_len` ini setting.
(same suffix as `log_errors_max_len`)
Allow values between 0 and 1000000 bytes.
For example, with zend.exception_string_param_max_len=0,
"" would represent the empty string, and "..." would represent something
longer than the empty string.
Previously, this was hardcoded as exactly 15 bytes.
Discussion: https://externals.io/message/110717
Closes GH-5769
Replace EG(autoload_func) with a C level zend_autoload hook.
This avoids having to do one indirection through PHP function
calls. The need for EG(autoload_func) was a leftover from the
__autoload() implementation.
Additionally, drop special-casing of spl_autoload(), and instead
register it just like any other autoloading function. This fixes
bug #71236 as a side-effect.
Finally, change spl_autoload_functions() to always return an array.
The distinction between false and an empty array no longer makes
sense here.
Closes GH-5696.
This is a bit tricky: In this cases we have "namespace as", which
means that we will only recognize "namespace" as an identifier when
the lookahead token is already at the "as". This means that
zend_lex_tstring picks up the wrong identifier.
We solve this by actually assigning the identifier as the semantic
value on the parser stack -- as in almost all cases we will not
actually need the identifier, this is just an (offset, size)
reference, not a copy of the string.
Additionally, we need to teach the lexer feedback mechanism used
by tokenizer TOKEN_PARSE mode to apply feedback to something
other than the very last token. To that purpose we pass through
the token text and check the tokens in reverse order to find the
right one.
Closes GH-5668.
get_gc() implementations that need to explore heterogeneous data
currently work by computing how many GC entries they need,
allocating a buffer for that and storing it on the object. This
is inefficient and wastes memory, because the buffer is retained
after the GC run.
This commit adds an API for a single global GC buffer, which can
be reused by get_gc implementations (as only one get_gc call is
ever active at the same time). The GC buffer will automatically
grow during the GC run and be discarded at the end.
Aside from a few very specific syntax errors for which detailed exceptions are
thrown, generally PHP just emits the default error messages generated by bison on syntax
error. These messages are very uninformative; they just say "Unexpected ... at line ...".
This is most problematic with constructs which can span an arbitrary number of lines, such
as blocks of code delimited by { }, 'if' conditions delimited by ( ), and so on. If a closing
delimiter is missed, the block will run for the entire remainder of the source file (which
could be thousands of lines), and then at the end, a parse error will be thrown with the
dreaded words: "Unexpected end of file".
Therefore, track the positions of opening and closing delimiters and ensure that they match
up correctly. If any mismatch or missing delimiter is detected, immediately throw a parse
error which points the user to the offending line. This is best done in the *lexer* and not
in the parser.
Thanks to Nikita Popov and George Peter Banyard for suggesting improvements.
Fixes bug #79368.
Closes GH-5364.
Also generate a fatal error if a collision occurs in zend_compile.
This is not perfect, because collisions might still be introduced
via opcache, if one file is included multiple times during a request,
invalidate in the meantime and recompiled by different processes.
This still needs to be addressed, but this patch fixes the much
more common case of collisions occuring when opcache is not used.
Fixes bug #78903.
Instead of handling shebang lines by adjusting the file pointer in
individual SAPIs, move the handling into the lexer, where this is
both a lot simpler and more robust. Whether the shebang should be
skipped is controlled by CG(skip_shebang) -- we might want to do
that in more cases.
This fixed bugs #60677 and #78066.
