Compile a call into another module - #882
Merged
Merged
Conversation
The interpreter has always been handed every file the compiler checked. The lowering was handed one, so a program with a `use` in it was refused, which is most programs that do any real work: eight of the thirty-four corpus files were waiting on this and nothing else. A `DefId` is an index into one module's table and a `Span` is an offset into one file, so neither means anything on the other side. A body from another module is read with that module's resolutions and that module's types, swapped in around the call rather than looked up per read, which keeps every existing lookup written the way it was. Only what is reached is lowered. A module that ships thirty functions and is imported for one contributes one, which matters because the rest of it may use shapes this backend cannot compile and refusing a program over a function nobody called would be wrong. A callee's `where` clause is kept whatever the other side worked out about its own callers. The check is dropped when every recorded call proved the clause, and the calls that reach an imported function were answered for in the caller's table, which the callee's module never saw. One thing came along with it. A type parameter that only appears inside a callback's type cannot be read off an argument, because a function value reaches this layer as one shape whatever it takes and hands back, and `map`'s `B` is exactly that. What the call itself came out as says it, so the return type is read when the arguments do not settle everything. Three more corpus files: json, tic_tac_toe, using_list. Twenty-two of thirty-four. Part of #877.
onatozmenn
added a commit
that referenced
this pull request
Aug 2, 2026
Follow-up to #882. A function reached across a module boundary can call its own module's functions, and those have no index in the program being compiled either. The lowering only looked for an import, so `markdown.deed` got refused at `slice`, which `std/string`'s own `replace` calls and nothing imports. Same path as an import: the callee is lowered from the module being read, as it is reached, into the same program. Nothing here is on the path of a program that imports nothing. The agreement test grew a function in the library that calls another one in the same library, since that was the case that was missing. Twenty-eight of thirty-four. Part of #877. <sub>Written with AI assistance.</sub>
onatozmenn
added a commit
that referenced
this pull request
Aug 2, 2026
Follow-up to #882. A call could cross a module boundary and a type could not, so a program that imported a record, a choice, an effect or a handler was refused however plain the call using it was. A type crosses the same way a function does, and the thing that matters is that it comes out as what the module that declared it built. Two layouts for one record would make a value of it fit neither, so each module's tables gain the names its `use` lines asked for rather than a copy of the shape. By name and only for what was asked for, so two modules that both declare a `Point` keep their own. The checker carries an imported type by module and name rather than by definition, for the same reason a `DefId` can't cross at all, and the lowering had no arm for that shape. It has one now, and the name is enough because the borrowing above put the right layout under it. The agreement test reads a field of a record the other module built and matches on a variant of a choice the other module declared, which is what would break if the two sides disagreed on the layout. Part of #877. <sub>Written with AI assistance.</sub>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
The interpreter has always been handed every file the compiler checked. The lowering was handed one, so a program with a
usein it was refused, which is most programs that do any real work: eight of the thirty-four corpus files were waiting on this and nothing else.The reason it wasn't just plumbing is that a
DefIdis an index into one module's table and aSpanis an offset into one file, so neither means anything on the other side. A body from another module gets read with that module's resolutions and that module's types, swapped in around the call rather than looked up per read, which keeps every existing lookup written the way it was.Two decisions I'd want a reviewer to look at:
Only what is reached is lowered. A module that ships thirty functions and is imported for one contributes one. That matters because the rest of it may use shapes this backend can't compile, and refusing a program over a function nobody called would be wrong.
A callee's
whereclause is kept whatever the other side worked out about its own callers. The check gets dropped when every recorded call proved the clause, and the calls that reach an imported function were answered for in the caller's table, which the callee's module never saw. There's a test for that one.One thing came along with it: a type parameter that only appears inside a callback's type can't be read off an argument, because a function value reaches this layer as one shape whatever it takes and hands back, and
map'sBis exactly that. What the call itself came out as says it, so the return type is read when the arguments don't settle everything.Three more corpus files compile:
json,tic_tac_toe,using_list. Twenty-two of thirty-four.Part of #877.
Written with AI assistance.