Const variables in prototypes don't add much value, but make the code
larger and redundant. Remove these const keywords.
Note, this doesn't apply to pointers to const memory.
ctx.c contains a workaround for systems without va_copy() support.
Improve this workaround in the way which is described in the autoconf
manual.
Also check for va_copy declaration in configure time.
Fixes#160.
There are situations when applications need more predictable timed
handlers that don't depend on connection status. Other usecase of
the global handlers is to manage offline connections, for example,
reconnect after a disconnection.
Introduced UNUSED macro with cast to void in commoh.h for internal
use. Used cast to void directly in those files which do not
include common.h. Although this change doesn't fix semantic issues
with unused function parameters, it does explicitly mark all those
places, which might require attention in future.
@sjaeckel integrated clang-format with formal coding style. Run his
script and commit changes.
There are pros and cons of this commit.
Mixed coding style is a "broken window". A good single style simplifies
reading and writing code.
On the other hand, this is a big change which will lead to conflicts.
Default timeout was 1 millisecond and it consumed CPU by running the
loop in xmpp_run(). The CPU usage was unacceptable and thus xmpp_run()
was not used by users. Also user can set the timeout by
xmpp_ctx_set_timeout() now.
Event loop already handles the situation when there is a timed handler
which needs to be run before the timeout is reached. Also system awakes
event loop when data arrives to socket.
The only drawback is when timed handler sends data it will be actually
sent after select(2) returns. But this situation can be handled easily by
adding socket to wfds when xmpp_conn_t contains data in sending queue.
This pseudo-random number generator solves problem with
platform-independent generation of randomized nonces.
Current implementation uses weak entropy, especially when
kernel.randomize_va_space = 0. But it can be improved by
adding new sources to xmpp_rand_reseed().
New internal API introduced:
xmpp_rand_new
xmpp_rand_free
xmpp_rand
xmpp_rand_bytes
xmpp_rand_nonce
Also fixed 2 bugs:
- only 1022 bytes of the message is printed when length is 1023.
- va_end() must be called after va_copy() otherwise we get
undefined behaviour according to C99.
va_end() has to be called after each invocation of va_copy(). So when
log messages are short, the copy was leaked. When log messages were
long, the old ap would get lost.
problems between Win32 (with its broken implementation) and the standard
stdio API. Also fixed a bug in debug output where long messages got
truncated to 1023 chars, and in xmpp_stanza_to_text where long stanzas
got truncated by one character.
(a filtered version of) the default logger which we implemented in
the library but was no longer accessible. This makes it much easier
for the caller to turn logging on by just passing this default
log handler callback pointer into xmpp_ctx_new() instead of having
to always implement boilerplate.
Another option would have been xmpp_get_default_log_level(ctx, level)
to change the filter level of the build in logger, which is installed
if you pass NULL to xmpp_ctx_new(). In this method the logger would
filter everything by default.
Defining static data in the header causes warnings on Apple gcc, and in general
this should be the client's responsibility. The names are obvious from the
log_level enum, which is public, and it's only one line of boilerplate to
copy in implementing a custom logging function.
