While running these on HHVM I've run into a lot of parallelism issues.
I'm backporting all the fixes I had to do in
https://github.com/facebook/hiphop-php/blob/master/hphp/tools/import_zend_test.py#L650
to php core.
Most of these changes were just filenames that were shared between
tests, but I did more surgery on the fixed ports. I can apreciate port
31337 as much as the next nerd, but random ports are better for tests.
* slim-postdata:
slim post data
add NEWS entry; add simple test
more precise condition
make this work in vc11 too
Use int64_t and atoll() after discussion with johannes
ws
Patch for https://bugs.php.net/bug.php?id=44522 to allow uploading files above 2G.
As noted in FR #65634, at present we don't send a Connection request header
when the protocol version is set to 1.1, which means that RFC-compliant Web
servers should respond with keep-alive connections. Since there's no way of
reusing the HTTP connection at present, this simply means that PHP will appear
to hang until the remote server hits its connection timeout, which may be quite
some time.
This commit sends a "Connection: close" header by default when HTTP 1.1 (or
later) is requested by the user via the context options. It can be overridden
by specifying a Connection header in the context options. It isn't possible to
disable sending of the Connection header, but given "Connection: keep-alive" is
the same as the default HTTP 1.1 behaviour, I don't see this as a significant
issue — users who want to opt in for that still can.
As a note, although I've removed an efree(protocol_version), this doesn't
result in a memory leak: protocol_version is freed in the out: block at the end
of the function anyway, and there are no returns between the removed efree()
and the later call. Yes, I ran the tests with valgrind to check that. ☺
Implements FR #65634 (HTTP wrapper is very slow with protocol_version 1.1).
Allows users to assert that something other than the backslash
should be considered an escape char; also follows the RFC 4180
recommendation that fields containing a " be enclosed.
This fixes issues that were found during static analysis by cjones where failure was impossible to detect due to return type mangling (casting an int to a char, then comparing to an int).
