Refactor julia initialization by finding and loading the sysimage earlier #58231
Conversation
| jl_gc_notify_image_load(sysimage.data, sysimage.size); | ||
|
|
||
| if (jl_options.cpu_target == NULL) | ||
| jl_options.cpu_target = "native"; |
There was a problem hiding this comment.
Should this be the default value in jloptions.c?
|
Is this enough to fix |
| @@ -461,9 +461,7 @@ JL_DLLEXPORT jl_gcframe_t **jl_autoinit_and_adopt_thread(void) | |||
| " (this should not happen, please file a bug report)\n"); | |||
| exit(1); | |||
| } | |||
| const char *image_path = jl_pathname_for_handle(handle); | |||
| jl_enter_threaded_region(); // This should maybe be behind a lock, but it's harmless if done twice | |||
There was a problem hiding this comment.
Was this deletion intentional?
There was a problem hiding this comment.
Yes. It got moved into the jl_init_with_handle call (the threaded region maybe should still be here :\ but I dunno)
There was a problem hiding this comment.
Yeah it was specifically the thread region I was curious about - I'm OK with deleting it for now, we can re-visit soon with thread-safety
topolarity
force-pushed
the
gb/refactor2
branch
from
2b4f4e0 to
7510499
Compare
This makes it clearer that sysimage resolution and `jl_options` setup are the responsibility of `jl_api.c`, not the initialization system.
topolarity
force-pushed
the
gb/refactor2
branch
from
7510499 to
8522727
Compare
topolarity
force-pushed
the
gb/refactor2
branch
from
7c9cb68 to
a242ff2
Compare
|
I cannot abide CI any longer. I'm going to merge this as-is (it has passed 80% of CI several times now) Pkg server issues have been worked around, but 5-10+ manual Windows restarts per run, random failures, and |
|
Thanks @gbaraldi ! |
This changes the order of initialization slightly. The goal is to be able to initialize executables and share libraries without having to explicitly call init functions.
This still doesn't address the thread safety