guys, always remember that every function that *generates output* could cause a
bailout if ignore_user_abort is set to false (and the user _aborts_ the
connection). in this case a longjump will be performed and our function (in
this case readfile) will have no chance to clean-up. having said that it's a
good idea to register all opened files using REGISTER_RESOURCE - that way the
engine will make sure they get closed on request end.
with an empty string as delimiter
@- Made fopen() of HTTP URL follow redirects, $http_response_header will
@ contain all headers with empty string as delimiter (Stig Venaas)
# Although man page for tzset() says it should be called automaticly
# by some of the time related functions, IRL this doesn't happen.
# PR: 3977, 4732, 5510, 6972, 7203, 8502, 8609
that all bytes in the character array have been set (they are used
to compute the hash value using hashpjw).
The function assumes that sprintf's %p modifier would always prefix
the output with "0x". On HPUX, this is not the case. Hence, not
all bytes may be properly initialized before being read.
This has been addressed by using only initialized bytes as the key.
Changed extract() to check that prefixed name is a valid one.
Optimized extract() a bit too.
# It'd be nice if someone made regression tests for extract().
- Make sapi_module available to external modules (PHPAPI)
- Make the php.ini path reported in phpinfo() always point to
real full path of the php.ini file
- Optimized the ISAPI module not to read unnecessary server
variables and read necessary variables at most once.
and array_diff(). Now using SORT_STRING instead of SORT_REGULAR
@- Fixed misbehavior of array_unique(), array_intersect() and array_diff();
@ could in certain cases give wrong result or segfault (Stig Venaas)
and then simply write it out to the target file. That avoids switching
between user and kernel land too many times. If that does not work, we
fall back to the read/write method.
