strongSwan

Noam,

The attached patch performs thread safe initialization, as opposed to the current initialization which is potentially unsafe

Thanks for your patch. I agree, that should be fixed.

Directly relying on <pthread.h> and pthread_once() is not ideal, though. Not all builds use pthreads, on Windows we use native threading primitives.

As we currently have no abstraction for pthread_once functionality (#640), for the time being we probably just should use explicit initialization.

I've pushed the referenced commit, queued for a merge to master.

Regards
Martin

For reference, pthread_once is used elsewhere in strongswan, so the proposed patch was not introducing a new dependency.

libstrongswan/threading/rwlock.c
libstrongswan/utils/chunk.c
libstrongswan/utils/utils/strerror.c
libfast/fast_request.c

This was true for releases before 5.2.0, but we explicitly removed these dependencies for the Windows support introduced with 5.2.0.

threading/rwlock.c is the pthread implementation of rwlocks, so that doesn't hurt. libfast is a very specific library not used in most builds, and has other Unix dependencies.

Regards
Martin