On x86 and x86_64, this change has no effect. On IA64, this fixes a
potential hardware exception. A function returned a partially initialised
value of aggregate type. The only caller of this function checks that the
value is valid before accessing it by testing the one definitely
initialised member. Therefore on x86 and derived architectures, there is
no uninitialised memory access. On IA64, with the regular calling
convention, the struct is allocated on the caller stack and passed as a
pointer, so there the uninitialised memory is also never accessed.
However, on calling conventions where one or more struct members past the
first byte are passed in registers or copied in memory, this call can
cause undefined behaviour.
Specifically, the value can contain a trap representation of the integers
(at the very least the 16 bit port) and cause a hardware exception and
SIGFPE in userland.
Regardless of the explanation above, this change fixes an instance of
undefined behaviour that just happened to be OK on all systems we tested
on.
Previously, the `ipv6` variable was initialised to `-1`, but that value
was never read. Either it was set to 0 or 1, or the function would return
before it was read. Thus, I've changed it to uninitialised `bool is_ipv4`
(inverted semantics to avoid negative conditions `if (!is_ipv6)`).
The `pack_ip_port` function is a bit unfortunate, but I'm not rewriting
it until we have a binary writer (I'd be rewriting it twice).
Reverted, since apparently cmake is not supposed to be used for things
other than testing/development and causes client build failures.
Apparently making it work for clients would require complicating
maintenance, which clearly can't be done.
This reverts commit 48ddb11599.
- CFLAG gnu99 was changed to c99.
- CXXFLAG c++98 was changed to c++11.
- CFLAG -pedantic-errors was added so that non-ISO C now throws errors.
- _XOPEN_SOURCE feature test macro added and set to 600 to expose SUSv3
and c99 definitions in modules that required them.
- Fixed tests (and bootstrap daemon logging) that were failing due to
the altered build flags.
- Avoid string suffix misinterpretation; explicit narrowing conversion.
- Misc. additions to .gitignore to make sure build artifacts don't wind
up in version control.
This was just for finding it in toktok-stack, which now uses
haskell-stack, and thus no longer has a .cabal-sandbox. We'll just assume
that the Makefile properly sets up the path such that tox-spectest is
available.
The CMAKE_SOURCE_DIR variable points to the top-level directory of a project while we want to point to the top-level directory of the lib. This change simplifies the integration as a third-party lib. We can use CMAKE_CURRENT_SOURCE_DIR or simply "nothing" ('.') to point to the directory where the CMakeLists.txt file resides.
Also added a `tox_options_copy` function for cloning an options object.
This can be useful when creating several Tox instances with slightly
varying options.
calls msi_kill() (toxav.c:180) which frees msi_call instances (msi.c:161)
which are then used when call_remove() (toxav.c:1136) is called.
This fix prevents call_remove() from calling invalid pointer.
Fixes#278
We create and destroy 20k tox instances and run a single tox_iterate on
it. This test is not comprehensive, but provides a simple check to see
whether the destruction properly cleans up memory and perhaps other
resources.
We really want to get all clients off this struct. We won't actually
remove it until 0.2, but we're going to break ABI compatibility with this
in various 0.1.x releases.
