Try to get GC analysis clean again #32090
Conversation
```
/home/keno/julia-old/src/gf.c:1220:30: note: Started tracking value here
jl_value_t *isect2 = jl_type_intersection((jl_value_t*)mth->sig, (jl_value_t*)sig);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/keno/julia-old/src/gf.c:1223:13: note: Passing non-rooted value as argument to function that may GC
if (jl_types_equal(isect, isect2)) {
^ ~~~~~~
```
|
@staticfloat Once this is merged, can we set up a separate buildbot that runs these? Perhaps it can also run the llvmpasses tests and later also the ClangSA test suite once that's hooked up. |
|
Yes; we already have a |
|
Not, I want it to run on every commit, it can just run in parallel with the regular tester. |
src/jltypes.c
Outdated
| @@ -1677,7 +1680,7 @@ void jl_init_types(void) JL_GC_DISABLED | |||
| // create base objects | |||
| jl_datatype_type = jl_new_uninitialized_datatype(); | |||
| jl_set_typeof(jl_datatype_type, jl_datatype_type); | |||
| jl_typename_type = jl_new_uninitialized_datatype(); | |||
| jl_typename_type = jl_new_uninitialized_datatype(); | |||
This gets everything except for rtutils.c (which is also moved out of the make target). rtutils will require some refactoring since the printing functions now are syntactically safe points (just happen not to be on the path we care about there).
| @@ -729,8 +730,10 @@ static char *jl_alloc_fiber(jl_ucontext_t *t, size_t *ssize, jl_task_t *owner) | |||
| char *stkbuf = (char*)jl_malloc_stack(ssize, owner); | |||
| if (stkbuf == NULL) | |||
| return NULL; | |||
| #ifndef __clang_analyzer__ | |||
There was a problem hiding this comment.
Curious why this is here?
There was a problem hiding this comment.
The analyzer gets confused by this syntax construct because it's inside a rooted value. &ptls->base_ctx. Could probably be fixed, but didn't seem worth the effort.
| @@ -44,9 +44,9 @@ RUNTIME_C_SRCS := \ | |||
| dlload sys init task array dump staticdata toplevel jl_uv datatype \ | |||
| simplevector runtime_intrinsics precompile \ | |||
| threading partr stackwalk gc gc-debug gc-pages gc-stacks method \ | |||
| jlapi signal-handling safepoint jloptions timing subtype rtutils \ | |||
| jlapi signal-handling safepoint jloptions timing subtype \ | |||
There was a problem hiding this comment.
Does this mean this file doesn't get analyzed now?
There was a problem hiding this comment.
yes, see commit message. jl_printf is now a safepoint, so that file needs some refactoring that I thought would be too extensive for this commit.
There was a problem hiding this comment.
Why is jl_printf now a safepoint?
There was a problem hiding this comment.
The mutex around uv_write can cause gc.
Also fix one missing GC root I found in the process.