Tim Hentenaar [Thu, 14 Jul 2016 13:45:01 +0000 (15:45 +0200)]
poll: Prevent libevent from spinning if POLLNVAL occurs
This can happen, for example if libevent is being used to poll fds given
by another library where the other library closes the fds without
notifying the program using it that said fds were closed. In this case,
libevent will simply spin on poll() since there are active fds, but
won't call any event callback to handle the condition.
In epoll case after socket closed it automatically removed from epfd, so
IOW it will not spin in epoll* API, just a timeout.
Azat Khuzhin [Tue, 1 Nov 2016 20:33:11 +0000 (23:33 +0300)]
Fix cmake -DEVENT__COVERAGE=ON
- do not use compiler check from the root cmake rules with syntax error,
let CodeCoverage check it
- fix CodeCoverage to check C compiler not CXX
- case insensitive checking of CMAKE_BUILD_TYPE
- replace flags with --coverage, and fix flags with linking with
--coverate, otherwise it will not compile during checking flags and
fails.
Azat Khuzhin [Mon, 31 Oct 2016 09:01:20 +0000 (12:01 +0300)]
autogen.sh: remove all autoconf/automake caches, if any
Otherwise if you for example share root of libevent repository with some
vm (for example windows+cygwin) then if you try to build with configure
on host, and then on guest it will fail, so let's just ignore this since
autogen.sh is just for this -- cleaning all stuff.
Zonr Chang [Wed, 24 Aug 2016 08:19:01 +0000 (16:19 +0800)]
cmake: cleanup
- Remove a redundant check on netdb.h (EVENT__HAVE_NETDB_H).
- Properly setup EVENT__SIZEOF_SSIZE_T for the case where "ssize_t" is
not "int".
- Remove unused EVENT__HAVE_PTHREAD.
- Set EVENT__HAVE_LIBZ instead of EVENT__HAVE_ZLIB (and remove the
latter).
Azat Khuzhin [Tue, 25 Oct 2016 21:51:56 +0000 (00:51 +0300)]
cmake/win32: fix running regress, but fixing finding python2 interpreter
Could not find executable C:/projects/libevent/build/bin/regress
Looked in the following places:
C:/projects/libevent/build/bin/regress
C:/projects/libevent/build/bin/regress.exe
...
9/13 Test #9: regress__WIN32_debug ................***Not Run 0.00 sec
Kurt Roeckx [Mon, 19 Sep 2016 20:05:15 +0000 (22:05 +0200)]
Make it build using OpenSSL 1.1.0
Rebased (azat):
- tabs instead of whitespaces
- make openssl-compat.h safe for complex expressions
- do not call sk_SSL_COMP_free() in 1.1 (fixes double free)
Adam Langley [Thu, 13 Oct 2016 00:49:17 +0000 (17:49 -0700)]
Don't call BIO_number_{read|written} on NULL BIOs.
OpenSSL doesn't document the behaviour of these functions when given a
NULL BIO, and it happens to return zero at the moment. But don't depend
on that.
Adding option to ignore clock_gettime: --disable-clock-gettime
macOS 10.12 introduced `clock_gettime` to libsystem. This means, built
on OS X 10.12 application would crash on earlier versions of OS X
because it will try to call clock_gettime. This options is useful to
make backwards compatible macOS apps.
Zonr Chang [Wed, 24 Aug 2016 09:16:32 +0000 (17:16 +0800)]
Fix evhttp_uriencode() regression.
http_uriencode_test() (in test/regress_http.c) has been failed after 72afe4c as "hello\0world" is encoded to "hello" instead of
"hello%00world". This is because of a misplaced overflow check which
causes the non-negative "size" specified in parameter being ignored in
within-bound URI.
Azat Khuzhin [Thu, 11 Aug 2016 13:15:45 +0000 (16:15 +0300)]
tests: use waitpid(..., WNOWAIT) to fix failing of main/fork under solaris
According to solaris docs:
"One instance of a SIGCHLD signal is queued for each child process whose
status has changed. If waitpid() returns because the status of a child
process is available, and WNOWAIT was not specified in options, any pending
SIGCHLD signal associated with the process ID of that child process is
discarded. Any other pending SIGCHLD signals remain pending."
And interesting thing that it works if you add sleep(1) before waitpid(), and
also if you run with --verbose (some race or what).
But linux doesn't support WNOWAIT in waitpid() so add detection into
cmake/autotools.
Azat Khuzhin [Thu, 11 Aug 2016 10:43:14 +0000 (13:43 +0300)]
test: replace sleeping with syncing pair in main/fork
About this syncing pair:
- read endpoint, must be blocked, to make it a checkpoint or smth like this
- write endpoint, must be nonblocking, to avoid readcb hung
During testing sometimes bufferevent/bufferevent_connect_fail_eventcb fails,
and after some digging I found that this was the case when connect() doesn't
return ECONNREFUSED immediately, and instead next operation will return it (in
our case evbuffer_read()->readv(), needless to say that after this
bufferevent_writecb() called and it checks error with getsockopt() but of
course it doesn't return any error), so this patch checks "errno" after
"readv()" and installs "bufev::connection_refused" flag, to handle this from
writecb and only once.
Azat Khuzhin [Tue, 9 Aug 2016 12:49:01 +0000 (15:49 +0300)]
Merge branch 'appveyor-cmake-v6'
This patch set introduces matrix for appveyor and fixes bugs related to win32,
during debugging this matrix. This should greatly reduce number of work for
making a release ('check under windows' part only).
Right now tests on appveyor still failes in summary, but we should dig into
problems, not hide them.
Test matrix looks like this:
- autotools
- autotools --disable-openssl
- autotools --disable-thread-support
- autotools --disable-debug-mode
- autotools --disable-malloc-replacement
- cmake
- cmake -DEVENT__DISABLE_OPENSSL=ON
- cmake -DEVENT__DISABLE_THREAD_SUPPORT=ON
- cmake -DEVENT__DISABLE_DEBUG_MODE=ON
- cmake -DEVENT__DISABLE_MM_REPLACEMENT=ON
- cmake -DEVENT__ENABLE_VERBOSE_DEBUG=ON
- cmake -DCMAKE_C_FLAGS='-DUNICODE -D_UNICODE'
And takes 90min (was 7min, and this is the sad part).
* appveyor-cmake-v6:
test/ssl: use send()/recv()/EVUTIL_ERR_RW_RETRIABLE()/EVUTIL_SOCKET_ERROR() to fix win32
test/https_basic: increase timeout for complete write (fixes win32)
cmake: check for ZLIB_INCLUDE_DIR, since we can have only library without headers
autotools/win32: fix searching ssl library
appveyor/autotools: link with openssl by passing LDFLAGS/CFLAGS
appveyor: image already had openssl installed
cmake/win32: do not compile regress_thread on -DEVENT__DISABLE_THREAD_SUPPORT=ON
cmake/win32: do not compile evthread_win32 on -DEVENT__DISABLE_THREAD_SUPPORT=ON
appveyor: check -DUNICODE -D_UNICODE according to ReleaseChecklist (cmake only)
cmake: fix -DEVENT__ENABLE_VERBOSE_DEBUG (typo on -DUSE_DEBUG)
test: fix building with --disable-thread-support under win32
cmake: do not use stderr for notifications/version-info
autoconf: fix --disable-thread-support build under win32
appveyor: ignore failure of mingw-get
appveyor: drop shallow_clone, since we use tags for detecting version in cmake
appveyor: support cmake & autotools using build matrix (like travis-ci has)
Azat Khuzhin [Mon, 8 Aug 2016 14:25:11 +0000 (17:25 +0300)]
appveyor/autotools: link with openssl by passing LDFLAGS/CFLAGS
This never works, hence green tests in production. and now it works:
https://ci.appveyor.com/project/azat/libevent/build/2.1.5.100/job/eib4xqfb7d3ruo16
(look by /"evthread_win32.c")
Azat Khuzhin [Mon, 8 Aug 2016 12:34:57 +0000 (15:34 +0300)]
appveyor: image already had openssl installed
Fixes messages like "Version mismatch for openssl: compiled with 1000201f but
running with 1000208f" [1].
According to next link:
http://help.appveyor.com/discussions/questions/1132-openssl-installation-issues
Refs: appveyor/ci#576
[1]: https://ci.appveyor.com/project/azat/libevent/build/2.1.5.95/job/dcy9cfoj3rfo4gdh
Azat Khuzhin [Sun, 7 Aug 2016 20:07:44 +0000 (23:07 +0300)]
cmake: do not use stderr for notifications/version-info
For all except "STATUS" message() will write to stderr, and this also includes
case when you does not have <mode> at all. Plus for every message in stderr
powershell generates exception and make appveyor fails.
Azat Khuzhin [Sun, 7 Aug 2016 18:30:04 +0000 (21:30 +0300)]
appveyor: ignore failure of mingw-get
Since according to doc [1], appveyour will fail the build on every powershell's
exception:
"PowerShell script is considered successful if it finishes without exception."
Azat Khuzhin [Sun, 19 Jun 2016 11:15:41 +0000 (14:15 +0300)]
appveyor: support cmake & autotools using build matrix (like travis-ci has)
This will allow us to avoid possible compilation regressions, and also some
failures in different environments (for example because of different
sizeof(size_t)).
For example we can avoid next issues:
- #361
- #314
- #311
(And I'm pretty sure that this is not complete list)
And of course it reduce job for maintainers.
But it also have one downside, appveyor doesn't have such concurrency like
travis-ci, so it can take a while to finish the whole build, but we can resolve
this later, or reduce build matrix to only one for autotools and one for cmake.
Mark Mentovai [Fri, 8 Jul 2016 13:43:39 +0000 (09:43 -0400)]
Check for Mac OS X 10.4 kqueue bug properly
EV_ERROR is a bit in struct kevent::flags. Other bits may be set too.
Actually we have osx builds on travis-ci, but it uses osx 10.9.5, and we don't
have warnings there, since I guess that there is no OR'ing with previous flag
in case of error, while in 10.12 there is OR.
Eduardo Panisset [Fri, 17 Jun 2016 17:46:32 +0000 (10:46 -0700)]
be_filter: avoid data stuck under active watermarks
Suppose we have bufferevent filter attached to bufferevent socket.
Read high watermark for bufferevent filter is configured to 4096 bytes.
Socket receives 4343 bytes. Due to watermark, 4096 bytes are transferred
from socket input buffer to filter input buffer and 247 bytes are left
in bufferevent socket.
Suppose that no more data is received through socket.
At this point 247 bytes will sit forever in input buffer of bufferevent
socket.
The patch attached solves this issue registering read callback to
filter's input buffer if it reaches its read high water mark and data
was left in corresponding underlying's input buffer.
This read callback calls filter process input function as soon as filter
input buffer falls below its read high watermark and there still is data
left in underlying input buffer. Callback is deregistered as soon as
filter input buffer falls below its read high watermark.
David Paschich [Sun, 22 May 2016 04:05:11 +0000 (21:05 -0700)]
Fix bufferevent_pair to properly set BEV_EVENT_{READING,WRITING} on flush.
Here's some fun. From `bufferevent.h`:
```
#define BEV_EVENT_READING 0x01 /**< error encountered while reading */
#define BEV_EVENT_WRITING 0x02 /**< error encountered while writing */
```
And from `event.h`:
```
/** Wait for a socket or FD to become readable */
#define EV_READ 0x02
/** Wait for a socket or FD to become writeable */
#define EV_WRITE 0x04
```
Library users have to be very careful to get this right; it turns out, the
library itself got this wrong in the `bufferevent_pair` code. It appears that
in most of the code, only `BEV_EVENT_FINISHED` will indicate whether it's read
or write; on error or timeout, it appears that "both" is assumed and not set in
the callback. I read through all the other places where `BEV_EVENT_FINISHED` is
passed to an event callback; it appears that the pair code is the only spot
that got it wrong.
azat: add TT_FORK to avoid breaking clean env, and rebase commit message
(copied from #359) Fixes: #359
Azat Khuzhin [Tue, 7 Jun 2016 15:31:48 +0000 (18:31 +0300)]
buffer: evbuffer_add_buffer(): clean empty chains from destination buffer
@EMPanisset reported a problem (#358) with evbuffer_remove_buffer(), but
actually I think that the problem is in evbuffer_add_buffer() which introduces
this empty chain, all other callers (except evbuffer_prepend_buffer(), but it
doesn't have this problem though) should be safe.
And FWIW the only API that allows empty chains is evbuffer_add_reference(), and
we can add check there to avoid such issues, but for now I leaved this without
fixing, since I think that evbuffer_add_reference() with empty chains can be
used as a barrier (but this can be tricky).
Azat Khuzhin [Tue, 7 Jun 2016 11:56:32 +0000 (14:56 +0300)]
test/buffer: evbuffer_remove_buffer() with empty chains (evbuffer_add_buffer())
Using:
- evbuffer_add()
- evbuffer_add_buffer() -- the one that has problem
- evbuffer_add_reference() -- the only one that allows empty chains to be added
- evbuffer_remove_buffer()
Possible failures after this patch set (not always, IOW in some builds this
issues aren't real issues):
- some failures but mostly because of timing issues, must be fixed separately.
- https://travis-ci.org/azat/libevent/jobs/129430229 # on brew update
- https://travis-ci.org/azat/libevent/jobs/129430221 # some locking issues
* travis-ci-os-matrix-v2:
automake: define serial-tests only if automake have this option
test/automake: don't use paralell test harness (since automake 1.12)
travis-ci/osx: relink gcc/g++ instead of clang
travis-ci: enable multi-os mode (osx, linux)
travis-ci: increase matrix (--disable-foo)
travis-ci: adjust alignment
Azat Khuzhin [Wed, 11 May 2016 07:18:25 +0000 (10:18 +0300)]
test/automake: don't use paralell test harness (since automake 1.12)
Starting from automake 1.2 there is parallel test harness, that redirects all
output to some log, which serial-test doesn't do.
So in case of new runner we can get no output for 10 minutes, for example on my
desktop:
$ time make verify VERBOSE=1
PASS: test/test-script.sh
============================================================================
Testsuite summary for libevent 2.1.5-beta
============================================================================
# TOTAL: 1
# PASS: 1
# SKIP: 0
# XFAIL: 0
# FAIL: 0
# XPASS: 0
# ERROR: 0
============================================================================
real 25m31.735s
user 0m13.753s
sys 0m7.648s
And this means that this will fail on travis-ci, since it has timeout for 10
minutes. Sure we can use `travis wait 60` instead, but I think that it is
better to fix this by writing result to output, instead of hacking around, so
let's use serial-tests instead of parallel always.
And now it works on travis-ci under linux because it has automake 1.11 while
osx has at least 1.12.
projects / libevent / log