eina_error_msg_get() must return NULL if an incorrect error is provided.
The XSI strerror_r() returns EINVAL when an invalid error is passed to
it, so we can end the function here. If we kept on, we would have tested
against the 'unknown_prefix' ("Unknown error ") which is implementation
defined, and registered a new error when the invalid error message
didn't match the 'unknown_prefix'. This new error message would have
been returned, which is not what we expected.
This case arised on Mac OS X where the 'unkwown prefix' is
"Unknown error: " instead of "Unknown error ".
It fixes eina test suite on Mac OS X.
Mac OS X does not support POSIX unnamed semaphores, only named
semaphores, which are persistant IPC: when the program exits,
and if semaphores where not released, they stay forever...
All EFL programs were "leaking" a semaphore, due to how
eina_log_monitor manages its resources. Therefore, after building
EFL a lot (which run eolian_gen, eolian_cxx, elua, edje_cc, ...)
we were not able to create any semaphore...
Now, we get rid of these semaphores and use Mac OS X's own
semaphores. Code is less cumbersome, and we don't have any
disavantage of the named semaphores.
Fixes T4423
@fix
So glibc has decided that readdir_r is hard to use safely and deprecated it
this summer. They recommand to use readdir, which was in the past unsafe to
use in a multi thread scenario, but is now on most system (and all system
we care, including our own implementation in evil). It is basically safe
as long the same DIRP is not accessed from another thread. This is true in
our code base, so we are fine to go with this.
For further reading: https://lwn.net/Articles/696474/
So actually there is quite a big issue with semaphores
on OSX. We use (named) POSIX semaphores, but this was
a (my) mistake... I'll fix it later...
The real issue is that named semaphore are persistants:
when the program dies, it stays alive. This is pretty
bad with eina_debug_monitor because we create a semaphore
we never release, due to a wild thread...
This leak of semaphores went unnoticed before commit
4a40ff95de because the
name of the semaphore was unique per process, and
overriden when another process was launched. This
was very bad, but saved us from overflowing the
semaphore pool. It is now overflowed pretty fast when
building a lot EFL, because of Eolian that runs A LOT!
So that's one problem that still needs to be fixed,
by using OSX' own semaphores (see T4423).
Another big issue, which is now fixed is that the
buffer in which we generated the semaphore ID was
too small, and therefore we were reduced to one shared
semaphore for a whole process... This buffer has been
now set to 31 characters, which seems to be the maximum
length of a semaphore ID.
So now things are better, but still with a deadly issue.
As pointed out in the mailing list, it was introduced in this release,
so it's better to remove the symbol instead of deprecating it.
People should use ETIMEDOUT directly.
So, first, the wrong strerror_r() was detected on
Mac OS X. Instead of using a complex set of macros
to try to detect which strerror_r() to use, when
it is defined, let the autotools handle that clerverness
for us.
we have some duplication of errors between Eina_Error and errno.h,
however we should use Eina_Error to extend the traditional errno.h
system.
then change eina_error_msg_register() and
eina_error_msg_static_register() to return a magic bit to state the
number was registered, and on other functions test this bit in order
to operate on registered values, otherwise fallback to errno.h, such
as strerror().
It also deprecates 2 clear duplicated errors:
- EINA_ERROR_OUT_OF_MEMORY -> ENOMEM
- EINA_ERROR_TIMEOUT -> ETIMEDOUT
There are two details when using strerror():
- old behavior did not return strings for non-error, such as
"Success" or "Unknown error ${N}"
- thread-safety issues: since we must be thread safe, then use
strerror_r() and eina_stringshare_add() that value, keeping a hash
of cached values
So it may be used outside EO (eina error is what I have in mind).
I believe it doesn't need to be redefined in all EFL libs, especially
since it's not used on Windows yet.
This reverts commit d19cd4e63c.
This causes a SIGBUS error on OpenBSD when closing any application. As this
was safety patch only I will revert it for the 1.18 release and we can work
out what breaks OpenBSD here for 1.19
Fixes T4332
Without this I get errors from the clang compiler used on Travis for some OSX
builds:
../src/lib/eina/eina_inline_lock_posix.x:845:27: error: use of undeclared identifier 'getpid'
we don't give any debug or info if a module in a list fails to load.
we should at least offer debug info. do that. this fixes coverity
issue CID 1039687
fixes CID 1361219
if you had an env var a few gb in size or we had stuff in the last bit
of memory address space this might be an issue, but that won't happen.
@fix
fixes CID 1361220
in theory yes end minus start could be insanely huge or end be very
high in memory thus causing an overflow. this would have to be in the
last few bytes of memory space, so it never going to happen. and the
input from the env var has to be sane anyway as its user controlled.
@fix
the way eina does sempahores, they can NEVER be sensibly shared
cross-process portably. so enabling sharing is a pointless idea. in
fact some os's like openbsd check if the sempahore addres is in a
sharable mem region and deny init if it is not. on osx you dont use
shared memory but a sempahore name you share instead... and this is
not exposed thus it can't be shared either. if we did process sharable
semaphores we'd make shm segments and/or name them in a sharable way
were you can share the idenitifer of the shm segment and/or the offset
address or name from osx. but we don't, so making them
process-private is the right thing. sharable sempahores will need a
whole new api.
this also fixes osx naming to make the name pretty unguessable/private
and opened exclusive (or it fails) by using pid, sem counter, and 4
random numbers. it's not a security mechanism as the create will fail
if there is a clash. chances are low. we unlink before anyway. good
enough for osx for now.
@fix
Coverity reports that eina_safepointer_get returns a NULL promise here
(checked 20 out of 21 times). As eina_safepointer_get can return NULL,
we should check the validity of 'promise' here before trying to
derefernce it later.
Fixes Coverity CID1356625
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
there was a leak of elements on the itrator stack when the stack was
flushed still having some items on it, thus losing their pointers and
never freeing them.
@fix
I guess the overflow was badly handled. Fixing it by using
explicit int intermediate value.
Fixes CID 1356616 and 1356619:
Operands don't affect result
Logically dead code
so. on linux signals are delivered to the main process thread/loop.
thats' where signal handlers are set up and always run. this is sane.
it's predicatble. but of course this is not the same in bsd land.
there "just send the signal to any old thread and call the signal
handler there" seems to tbe the order of the day. this explains why
wer are losing sigchld signals in edje_cc - it's heavily threaded and
bsd is just randombly picking a thread to call it on.
this fixes that. in theory. i hope. i can't test, but putting it in to
share
@fix
This is a reccurring compatibility issue...
MAP_ANONYMOUS is not defined on OSX, but MAP_ANON is.
I know MAP_ANON is marked as deprecated in the Linux man
pages, but it has the benefit of being more portable.
This is heavily inspired from Eo_Id infrastructure. Main change
are that the lower bit are always guaranteed to be zero and ignored
by all function. Also it may be a little bit less efficient in some
case, but we will tune it once we have real life usage of it.
Eo won't be migrated for 1.18 to it as Eo_Id is deeply integrated
and it is quite risky to touch it so close from a freeze. This can
wait.
In case of failure within eina_lock_new() (posix), a
pthread_mutexattr_t would have been left in an initialized state,
without any deinitialization being called.
Consequences would have been implementation defined.
DragonFlyBSD has pthread_setname_np but no pthread_setaffinity_np;
we still need to include pthread_np though.
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
Jean Guyomarc'h
Cedric BAIL
Gustavo Sverzut Barbieri
Jean-Philippe Andre
Carsten Haitzler
Stefan Schmidt
Chris Michael
Jean Guyomarc'h
Felipe Magno de Almeida
Lauro Moura
Benjamin Jacobs
Larry Jr