Refactoring block_for_gc #100
Conversation
| // when executing finalizers do not let another thread do GC (set a variable such that while that variable is true, no GC can be done) | ||
| int8_t set_gc_initial_state(void* ptls_raw) | ||
| // based on jl_gc_collect from gc.c | ||
| JL_DLLEXPORT void jl_gc_prepare_to_collect(void) |
There was a problem hiding this comment.
Looks like this function just duplicates jl_gc_collect. Have you considered moving the function to gc-common.c, and letting Julia and MMTk each implement a different _jl_gc_collect?
There was a problem hiding this comment.
The problem is that triggering GC manually in Julia is done via a call to jl_gc_collect (https://github.com/mmtk/julia/blob/83796a7d38878faa9afdb94ce84b6da00035b336/base/gcutils.jl#L129). For MMTk, we want that function to call into mmtk_handle_user_collection_request as it currently does. That's why I did the duplication, with jl_gc_prepare_to_collect doing the actual preparation that is done in jl_gc_collect from gc.c.
We could potentially reuse jl_gc_collect for what I'm doing by passing a different argument (eg. -1), but we'd still have the duplication - the code for Julia in gc.c and the code for MMTk in mmtk-gc.c.
Currently, we keep calling into C multiple times to set up some variables before the thread stops for GC and after GC is done. This PR refactors the code in
block_for_gcto delegate that to C, reusing most of the code injl_gc_collect. That should help preventing errors in the future due to the mismatch between the implementations as it keeps things in a single place.Should be merged with mmtk/julia#29.