Ensure CI from external MI and foreign abs interpreter survive precompilation

[vchuravy]

• Improve precompile tests
• add test from GPUCompiler

Forward-port of #60741

[vchuravy]

I expect that we also need:

diff --git a/src/staticdata_utils.c b/src/staticdata_utils.c
index 16ebcd4b4ee..ad6aceb411e 100644
--- a/src/staticdata_utils.c
+++ b/src/staticdata_utils.c
@@ -502,7 +502,7 @@ static jl_array_t *queue_external_cis(jl_array_t *list, jl_query_cache *query_ca
         int dispatch_status = jl_atomic_load_relaxed(&m->dispatch_status);
         if (!(dispatch_status & METHOD_SIG_LATEST_WHICH))
             continue; // ignore replaced methods
-        if (ci->owner == jl_nothing && jl_atomic_load_relaxed(&ci->inferred) && jl_is_method(m) && jl_object_in_image((jl_value_t*)m->module)) {
+        if (jl_atomic_load_relaxed(&ci->inferred) && jl_is_method(m) && jl_object_in_image((jl_value_t*)m->module)) {
             int found = has_backedge_to_worklist(mi, &visited, &stack, query_cache);
             assert(found == 0 || found == 1 || found == 2);
             assert(stack.len == 0);

but I want to ensure that the new test fails.

[vchuravy]

@aviatesk I added a new test here instead of adding to precopile_absintX since the structure is slightly different.

The test comes from
https://github.com/JuliaGPU/GPUCompiler.jl/blob/master/test/native/precompile.jl

where we have a "compiler" package and a "user" package, and we are adding a code instance to a method instance that is already owned somewhere else identity(::Any) which is why we need the PrecompileTools-esque setup.

[vtjnash]

Something probably went horribly wrong if ci.max_world != WORLD_AGE_REVALIDATION_SENTINEL, since that means the Compiler made a mistake and forgot to get it validated

[vchuravy]

@vtjnash I am a bit confused about this query:

1. There could be more than one entry? We should select the one that is valid in the current world.

The values I am observing here are:

ci = CodeInstance for MethodInstance for ExampleUser.square(::Float64)
ci.max_world = 0xffffffffffffffff
ci.min_world = 0x0000000000009b8c
ci = CodeInstance for MethodInstance for identity(::Any)
ci.max_world = 0xffffffffffffffff
ci.min_world = 0x0000000000009b8c

Which does look to me like they are correctly integrated? But they of course fail the test, since max_world != WORLD_AGE_REVALIDATION_SENTINAL and min_world is not <=1

[vchuravy]

@MasonProtter given #60442 could you write some tests that check that foreign code instances that are part of a Expr(:invoke are handled correctly during precompilation? Otherwise we might generate code that segfaults later.