Use debug printing for expect_test in the parser and other places. #1321
Labels
Comments
sezna
added
bug
This was referenced Apr 6, 2024
|
@sezna , I would like to work on this issue. |
github-merge-queue bot
pushed a commit
that referenced
this issue
May 17, 2024
_This PR replaces #1364_ This PR introduces a feature called _hierarchical namespaces_, a module-like system that maintains backward compatibility. It is the foundational groundwork that will pave the way for: 1. Naked namespaces (files without namespace declarations) 2. Package publishing and sharing 3. Explicit item imports 4. Exports and re-exports 5. and the rest of the stuff [here](#493 (comment)) Key changes include: 1. `VecIdent` type introduction: This abstraction over `Vec<Ident>` represents the new namespaces in the AST. Alternate names could be `DotIdent` or `Path`, but those names were already loaded with meaning in our codebase. I'm open to feedback on nomenclature here. 7. `NamespaceId` usage: This replaces our previous flat map-style `Rc<str> -> Res` item tracking. 8. `NamespaceTree`: This tracks the namespace hierarchy separately in low-cost data structure, and allows us to keep our basic `tys` and `terms` maps relatively untouched. 9. Name resolution algorithm update: The algorithm in `resolve.rs` received a heavy facelift. Not only did I try to simplify the existing code, but I also added the required functionality to check candidate partially-opened or re-opened namespaces. This PR maintains 100% backward compatibility with the existing resolution engine, preserving features like re-opening namespaces, shadowing semantics, merging namespaces via aliases, and absolute-path opens. It also introduces the ability to reference a function in a different namespace if its parent namespace is opened. Additional notes: 1. Katas used a form of shadowing that wasn't tested in our compiler. This PR [adds a test](https://github.com/microsoft/qsharp/pull/1312/files#diff-7a3aa16f18f6d132b9e54a6b6ebd53e6c3b41cea805c8f8db9c456b533a3c96bR2787) for that. 2. Pursuing #1321 could help differentiate user-facing and compiler-engineer-facing string representations of structs, which are increasingly conflated. 3. There's potential for refactoring `ast`, `fir`, and `hir` to use shared generic types, reducing duplicate `Ident`/etc types. 4. Completions functionality remains the same, but it's not yet aware of the ability to partially open namespaces (e.g. `open Microsoft.Quantum`. This will be addressed in future updates. 5. There is a performance regression, but we are performing a full tree-traversal namespace resolution algorithm in a place where we used to just do a constant table lookup. I have optimized this as much as I believe is reasonable for now, bringing the regression down from ~40%, where it was originally, to around 5-10% based on the benchmark run. Also closes #1336 --------- Co-authored-by: Mine Starks <mineyalc@microsoft.com> Co-authored-by: Mine Starks <16928427+minestarks@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
We currently use
ToString::to_string(i.e.,Display) in our expect test tests for the parser (see here). This is problematic, because this trait is typically used for user-facing representations of data structures. We are forced to choose between a pretty user-facing representation, and a more technically accurate debug-style representation for testing for namespace names in #1312 as a result.The text was updated successfully, but these errors were encountered: