sdl-gnash built on Arch Linux fails to hook

[qixils]

Using a build of sdl-gnash from this custom PKGBUILD for gnash-git (fixes annoying/long dependencies + errors in the actual package)

libTAS attempts and fails to connect to socket. libTAS output/gdb backtrace -- warning: it's.. long

I noticed various clock_gettime calls and thought it might be a timing issue, but tweaking with timing settings along with all other (debug) options hasn't provided any different results

[clementgallet]

There's a recursive call here caused by jemalloc:
je_malloc()
-> libtas::clock_gettime() (for some arena profiling)
-> dlsym() (to fetch the real clock_gettime() function pointer)
-> calloc() (dlsym() allocates the dlerror() string buffer on first call)
-> je_malloc().

[clementgallet]

It took me some time to figure out where to break the recursive call. Working on calloc() was a deadend, so worked on dlsym(). When a recursive dlsym() call is detected, we don't call the original function, but the internal implementation-dependant _dl_sym(), which bypasses all side effects of the API function, avoiding further recursive calls. It's far from perfect, but it works. Also, if _dl_sym() trick breaks, we have bigger problems than this :/