Since long the default PHP charset is UTF-8, however the Windows part is
out of step with this important point. The current implementation in PHP
doesn't technically permit to handle UTF-8 filepath and several other
things. Till now, only the ANSI compatible APIs are being used. Here is more
about it
https://msdn.microsoft.com/en-us/library/windows/desktop/dd317752%28v=vs.85%29.aspx
The patch fixes not only issues with multibyte filenames under
incompatible codepages, but indirectly also issues with some other multibyte
encodings like BIG5, Shift-JIS, etc. by providing a clean way to access
filenames in UTF-8. Below is a small list of issues from the bug tracker,
that are getting fixed:
https://bugs.php.net/63401https://bugs.php.net/41199https://bugs.php.net/50203https://bugs.php.net/71509https://bugs.php.net/64699https://bugs.php.net/64506https://bugs.php.net/30195https://bugs.php.net/65358https://bugs.php.net/61315https://bugs.php.net/70943https://bugs.php.net/70903https://bugs.php.net/63593https://bugs.php.net/54977https://bugs.php.net/54028https://bugs.php.net/43148https://bugs.php.net/30730https://bugs.php.net/33350https://bugs.php.net/35300https://bugs.php.net/46990https://bugs.php.net/61309https://bugs.php.net/69333https://bugs.php.net/45517https://bugs.php.net/70551https://bugs.php.net/50197https://bugs.php.net/72200https://bugs.php.net/37672
Yet more related tickets can for sure be found - on bugs.php.net, Stackoverflow
and Github. Some of the bugs are pretty recent, some descend to early
2000th, but the user comments in there last even till today. Just for example,
bug #30195 was opened in 2004, the latest comment in there was made in 2014. It
is certain, that these bugs descend not only to pure PHP use cases, but get also
redirected from the popular PHP based projects. Given the modern systems (and
those supported by PHP) are always based on NTFS, there is no excuse to keep
these issues unresolved.
The internalization approach on Windows is in many ways different from
UNIX and Linux, while it supports and is based on Unicode. It depends on the
current system code page, APIs used and exact kind how the binary was compiled
The locale doesn't affect the way Unicode or ANSI API work. PHP in particular
is being compiled without _UNICODE defined and this is conditioned by the
way we handle strings. Here is more about it
https://msdn.microsoft.com/en-us/library/tsbaswba.aspx
However, with any system code page ANSI functions automatically convert
paths to UTF-16. Paths in some encodings incompatible with the
current system code page, won't work correctly with ANSI APIs. PHP
till now only uses the ANSI Windows APIs.
For example, on a system with the current code page 1252, the paths
in cp1252 are supported and transparently converted to UTF-16 by the
ANSI functions. Once one wants to handle a filepath encoded with cp932 on
that particular system, an ANSI or a POSIX compatible function used in
PHP will produce an erroneous result. When trying to convert that cp932 path
to UTF-8 and passing to the ANSI functions, an ANSI function would
likely interpret the UTF-8 string as some string in the current code page and
create a filepath that represents every single byte of the UTF-8 string.
These behaviors are not only broken but also disregard the documented
INI settings.
This patch solves the issies with the multibyte paths on Windows by
intelligently enforcing the usage of the Unicode aware APIs. For
functions expect Unicode (fe CreateFileW, FindFirstFileW, etc.), arguments
will be converted to UTF-16 wide chars. For functions returning Unicode
aware data (fe GetCurrentDirectoryW, etc.), resulting wide string is
converted back to char's depending on the current PHP charset settings,
either to the current ANSI codepage (this is the behavior prior to this patch)
or to UTF-8 (the default behavior).
In a particular case, users might have to explicitly set
internal_encoding or default_charset, if filenames in ANSI codepage are
necessary. Current tests show no regressions and witness that this will be an
exotic case, the current default UTF-8 encoding is compatible with any
supported system. The dependency libraries are long switching to Unicode APIs,
so some tests were also added for extensions not directly related to streams.
At large, the patch brings over 150 related tests into the core. Those target
and was run on various environments with European, Asian, etc. codepages.
General PHP frameworks was tested and showed no regressions.
The impact on the current C code base is low, the most places affected
are the Windows only places in the three files tsrm_win32.c, zend_virtual_cwd.c
and plain_wrapper.c. The actual implementation of the most of the wide
char supporting functionality is in win32/ioutil.* and win32/codepage.*,
several low level functionsare extended in place to avoid reimplementation for
now. No performance impact was sighted. As previously mentioned, the ANSI APIs
used prior the patch perform Unicode conversions internally. Using the
Unicode APIs directly while doing custom conversions just retains the status
quo. The ways to optimize it are open (fe. by implementing caching for the
strings converted to wide variants).
The long path implementation is user transparent. If a path exceeds the
length of _MAX_PATH, it'll be automatically prefixed with \\?\. The MAXPATHLEN
is set to 2048 bytes.
Appreciation to Pierre Joye, Matt Ficken, @algo13 and others for tips, ideas
and testing.
Thanks.
gdb seems to be not always able to fetch the correct context for
thread locals. Thus, the "if (TSRMLS_CACHE)" clause causes gdb to
show crashes which aren't recognized neither with helgrind nor
in release builds. This is reproducable by setting breakpoints on
the exact line where PHP has a tsrm_get_ls_cache() call.
* origin/master: (398 commits)
NEWS
add test for bug #68381
Fixed bug #68381 Set FPM log level earlier during init
proper dllexport
move to size_t where zend_string is used internally
fix some datatype mismatches
return after the warning, to fix uninitialized salt usage
fix datatype mismatches
add missing type specifier
fix datatype mismatches
fix unsigned check
"extern" shouldn't be used for definitions
joined identical conditional blocks
simplify fpm tests
SEND_VAR_NO_REF optimization
Add test for bug #68442
Add various tests for FPM - covering recent bugs (68420, 68421, 68423, 68428) - for UDS - for ping and status URI - for multi pool and multi mode
Include small MIT FastCGI client library from https://github.com/adoy/PHP-FastCGI-Client
Get rid of zend_free_op structure (use zval* instead). Get rid of useless TSRMLS arguments.
Add new FPM test for IPv4/IPv6
...
Conflicts:
win32/build/config.w32
* origin/master:
properly export tsrm_strtok_r()
Fixed list() behavior inconsistency (string handling is disabled for all cases, ArrayAccess objects handling is enabled for all cases, ZEND_FETCH_DIM_TMP_VAR opcode is renamed into ZEND_FETCH_LIST, ZEND_FETCH_ADD_LOCK flag is removed).
Fix accidental edit in previous commit
Fix arginfo
DateTimeZone::getOffset() now accepts a DateTimeInterface
DateTimeZone::getOffset() now accepts a DateTimeInterface
Moved checks and error reporting related to static methods from DO_FCALL inti INTI_FCALL* opcodes that may really deal with static methods. (In some rare cases it may lead to different order of warning messages).
Fixed a bug that causes crash when environment variable is access while parsing php.ini
Fixed a bug that causes crash when environment variable is access while parsing php.ini
* origin/master:
remove the remains of dsp files handling
fix EX usage
remove misprint parentheses
remove misprint parentheses
Replaced EG(This) and EX(object) with EX(This). Internal functions now recieves zend_execute_data as the first argument.
And this one...
It should be in extern c
Remove useless condition
NEWS entry for previous commit
NEWS entry for previous commit
add IPv6 support to php-fpm
Micro optimization for the most frequency case
Add hash to EXTENSIONS file
Remove extensions which are long gone
we also have xz release tarballs since 5.5
Fix ZTS build
improved file size computation in stat()
Fixed incorrect compilation
5.5.19 now
* origin/master: (26 commits)
Micro optimization
Drop unused INIT_STRING opcode
Drop unused RAISE_ABSTRACT_ERROR opcode
CT substitute unqualified true/false/null in namespaces
Fix a couple compile warnings
fix test filename
one more test to illustrate transfer of an arbitrary data amount throug pipes
fix tests on linux
better test cleanup
Use more readable inline functions
increase the polling period to not to break existing behaviours
updated NEWS
Fixed bug #51800 proc_open on Windows hangs forever
Fixed segfault
Set an LDAP error code when failing ldap_bind due to null bytes
Fix segmentation fault in debug_backtrace()
Drop support for GMP 4.1
Make gmp_setbit and gmp_clrbit return values consistent
removed *.dsw and *.dsp files
Opcache compatibility for coalesce operator
...
TLS is already used in TSRM, the way exporting the tsrm cache through
a thread local variable is not portable. Additionally, the current
patch suffers from bugs which are hard to find, but prevent it to
be worky with apache. What is done here is mainly uses the idea
from the RFC patch, but
- __thread variable is removed
- offset math and declarations are removed
- extra macros and definitions are removed
What is done merely is
- use an inline function to access the tsrm cache. The function uses
the portable tsrm_tls_get macro which is cheap
- all the TSRM_* macros are set to placebo. Thus this opens the way
remove them later
Except that, the logic is old. TSRMLS_FETCH will have to be done once
per thread, then tsrm_get_ls_cache() can be used. Things seeming to be
worky are cli, cli server and apache. I also tried to enable bz2
shared and it has worked out of the box. The change is yet minimal
diffing to the current master bus is a worky start, IMHO. Though will
have to recheck the other previously done SAPIs - embed and cgi.
The offsets can be added to the tsrm_resource_type struct, then
it'll not be needed to declare them in the userspace. Even the
"done" member type can be changed to int16 or smaller, then adding
the offset as int16 will not change the struct size. As well on the
todo might be removing the hashed storage, thread_id != thread_id and
linked list logic in favour of the explicit TLS operations.
