Test failures on alpine linux about bdwgc HOT 6 CLOSED

To simplify debugging:

• use bdwgc master branch
• add parameters to configure: --enable-gc-assertions --disable-threads
• do not use "incremental mode" (-D GC_DISABLE_INCREMENTAL)
• focus on gctest (test.c) debugging (you could even compile it w/o configure: gcc -I include -D GC_ASSERTIONS -D GC_DISABLE_INCREMENTAL -O0 -g tests/test.c)

from bdwgc.

It seems gc currently never defines HAVE_DL_ITERATE_PHDR on non-glibc Linux platforms. Musl have the one and it is required to work with libc.so properly.

By adding it on dyn_load.c, --enable-gc-assertions --disable-threads passes make check.

For threading support, Musl's default stack size is way too small compared to glibc; at least 80k seems not enough to parallel markers. I think we need 256K-ish here.

I'm not sure why incremental collection is not working on Musl yet..

from bdwgc.

Please contribute the patch

from bdwgc.

The hard part is Musl have no __MUSL__ as the FAQ says so there is no way to statically add any Musl specific workarounds.

Possible solutions are...

1. Determine dl_iterate_phdr existence with AC_TRY_COMPILE
2. Consider defined(LINUX) && !defined(__GLIBC__) && !defined(__UCLIBC__) && !defined(__dietlibc__) as Musl
3. Just let distributors to do the workaround

I'm now trying to do 2 but just started research around dietlibc..(and pray there is no other libc exists on Linux)

In case try-compile preferred over 2, I will prepare patch against configure.ac and CMakeLists.txt but I believe we have historically avoided doing any dynamic configuration.

from bdwgc.

I suggest solution 2.
Related issues: #70 #84

from bdwgc.

Finally I implemented the 1st approach - see my comment in #148
I have not solved the issue with small stacks for parallel markers, so --disable-parallel-mark should be passed configure.

If you have some patches not submitted to upstream, you are welcome.

PS. Topics related to musl: #66, #70, #72, #84, #123, #148

from bdwgc.