Refactor generated function implementation to provide bindings/method #57230
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
14 tasks
Keno
force-pushed
the
kf/generatedrefactor
branch
3 times, most recently
from
018796a to
a016b2b
Compare
aviatesk
reviewed
Feb 2, 2025
Keno
force-pushed
the
kf/generatedrefactor
branch
6 times, most recently
from
3fb8912 to
62b3fcf
Compare
This PR refactors the generated function implementation in multiple ways: 1. Rather than allocating a new LineNumber node to pass to the generator, we just pass the original method from which this LineNumberNode was constructed. This has been a bit of a longer-standing annoyance of mine, since the generator needs to know properties of the original method to properly interpret the return value from the generator, but this information was only available on the C side. 2. Move the handling of `Expr` returns fully into Julia. Right not things were a bit split with the julia code post-processing an `Expr` return, but then handing it back to C for lowering. By moving it fully into Julia, we can keep the C-side interface simpler by always getting a `CodeInfo`. With these refactorings done, amend the post-processing code to provide binding edges for `Expr` returns. Ordinarily, bindings in lowered code do not need edges, because we will scan the lowered code of the method to find them. However, generated functions are different, because we do not in general have the lowered code available. To still give them binding edges, we simply scan through the post-lowered code and all of the bindings we find into the edges array. I will note that both of these will require minor adjustments to `@generated` functions that use the CodeInfo interface (N.B.: this interface is not considered stable and we've broken it in almost every release so far). In particular, the following adjustments need to be made: 1. Adjusting the `source` argument to the new `Method` ABI 2. If necessary, adding any edges that correspond to GlobalRefs used - the code will treat the returned CodeInfo mostly opaquely and (unlike in the `Expr` case) will not automatically compute these edges.
Keno
force-pushed
the
kf/generatedrefactor
branch
from
62b3fcf to
30c6ff9
Compare
vtjnash
reviewed
Feb 3, 2025
| JL_DLLEXPORT void jl_add_binding_backedge(jl_binding_t *b, jl_value_t *edge) | ||
| { | ||
| if (!b->backedges) { | ||
| b->backedges = jl_alloc_vec_any(0); |
There was a problem hiding this comment.
It looks like there is no lock on this, so any uses of ./julia (which now defaults to using threads) and which run any code now may trigger this UB memory corruption?
vtjnash
reviewed
Feb 3, 2025
| Expr(:meta, :pop_loc)))) | ||
| spnames = g.spnames | ||
| return generated_body_to_codeinfo(spnames === Core.svec() ? lam : Expr(Symbol("with-static-parameters"), lam, spnames...), | ||
| typename(typeof(g.gen)).module, |
There was a problem hiding this comment.
Why aren't we using source.module here anymore? I suppose they probably usually resolve to the same thing, but might be good to document the intent of this change anyways?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR refactors the generated function implementation in multiple ways:
Rather than allocating a new LineNumber node to pass to the generator, we just pass the original method from which this LineNumberNode was constructed. This has been a bit of a longer-standing annoyance of mine, since the generator needs to know properties of the original method to properly interpret the return value from the generator, but this information was only available on the C side.
Move the handling of
Exprreturns fully into Julia. Right not things were a bit split with the julia code post-processing anExprreturn, but then handing it back to C for lowering. By moving it fully into Julia, we can keep the C-side interface simpler by always getting aCodeInfo.With these refactorings done, amend the post-processing code to provide binding edges for
Exprreturns. Ordinarily, bindings in lowered code do not need edges, because we will scan the lowered code of the method to find them. However, generated functions are different, because we do not in general have the lowered code available. To still give them binding edges, we simply scan through the post-lowered code and all of the bindings we find into the edges array.I will note that both of these will require minor adjustments to
@generatedfunctions that use the CodeInfo interface (N.B.: this interface is not considered stable and we've broken it in almost every release so far). In particular, the following adjustments need to be made:sourceargument to the newMethodABIExprcase) will not automatically compute these edges.