Add support for C-Sharp

[dburriss]

I imagine we would use tree-sitter-c-sharp.

Since the Java metrics are growing I would like to add C# too. I imagine the implementations should be close to identical.

[marco-c]

Good idea!

[dburriss]

So folding in lessons learned from #694 I am starting with some research and definitions. Thankfully C# has a fairly clear definition of a statement as well as a comprehensive list of the different types:

csharp statements

A cursory look through seems to indicate, unsurprisingly, that it is much the same as Java. I will raise it if I find any weird difference.

Should I create an issue per metric to track this or link PRs here?

[dburriss]

So bumping into my first challenge here. On this repo currently tree-sitter version dependency istree-sitter = ">= 0.19, < 0.21".
The source Cargo.toml for tree-sitter-c-sharp is tree-sitter = "0.19".
On crates.io the tree-sitter version is 0.17.

I guess this is the reason I get:

expected struct tree_sitter::Language, found a different struct tree_sitter::Language

when trying to add tree-sitter-c-sharp as a dependency?

I assume the fix here is to ask @dcreager, the owner of the crate, if it can be updated?

[dcreager]

Just published tree-sitter-c-sharp 0.19.1, which should unblock you if you're using tree-sitter 0.19.

If you're using tree-sitter 0.20, then I think you'll end up getting the same error, until https://github.com/tree-sitter/tree-sitter-c-sharp/pull/219 (which I just opened) is merged. Once that gets a :+1: review I'll merge it and cut a new release of tree-sitter-c-sharp with the more inclusive tree-sitter dependency.

[dburriss]

Thanks so much @dcreager 🙏 I'll test it out tomorrow but I think it should work.

[marco-c]

Should I create an issue per metric to track this or link PRs here?

Let's just use this one, maybe you can edit the first comment to add a list of metrics with checkmarks.

[dburriss]

After adding tree_sitter_c_sharp to the nums crate I get the follwong compile error:

 Compiling enums v0.0.1 (C:\Users\devon\GitHub\rust-code-analysis\enums)
error[E0308]: mismatched types
  --> src\macros.rs:22:31
   |
20 |           pub fn get_language(lang: &Lang) -> Language {
   |                                               -------- expected `tree_sitter::Language` because of return type
21 |               match lang {
22 |                   Lang::Java => tree_sitter_java::language(),
   |                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected struct `tree_sitter::Language`, found a different struct `tree_sitter::Language`
   |
  ::: src\languages.rs:3:1
   |
3  | / mk_langs!(
4  | |     // 1) Name for enum
5  | |     // 2) tree-sitter function to call to get a Language
6  | |     (Java, tree_sitter_java),
...  |
15 | |     (Javascript, tree_sitter_javascript)
16 | | );
   | |_- in this macro invocation
   |
   = note: perhaps two different versions of crate `tree_sitter` are being used?
   = note: this error originates in the macro `mk_get_language` (in Nightly builds, run with -Z macro-backtrace for more info)

I am not 100% sure what is going on here. Running cargo lock -d, tree-sitter version look like so:

- tree-sitter 0.19.3
  - cc 1.0.72
  - regex 1.4.6
- tree-sitter 0.20.2
  - cc 1.0.72
  - regex 1.4.6
- tree-sitter-ccomment 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-cpp 0.20.0
  - cc 1.0.72
  - tree-sitter 0.20.2
- tree-sitter-java 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-javascript 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-mozcpp 0.20.1
  - cc 1.0.72
  - tree-sitter 0.19.3
  - tree-sitter-cpp 0.20.0
- tree-sitter-mozjs 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
  - tree-sitter-javascript 0.19.0
- tree-sitter-preproc 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-python 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-rust 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-typescript 0.20.0
  - cc 1.0.72
  - tree-sitter 0.19.3

Then when I add tree-sitter-c-sharp = "0.19.1" to enums/Cargo.toml, the dependencies look like so:

- tree-sitter 0.19.3
  - cc 1.0.72
  - regex 1.4.6
- tree-sitter 0.20.2
  - cc 1.0.72
  - regex 1.4.6
- tree-sitter-c-sharp 0.19.1
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-ccomment 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-cpp 0.20.0
  - cc 1.0.72
  - tree-sitter 0.20.2
- tree-sitter-java 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-javascript 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-mozcpp 0.20.1
  - cc 1.0.72
  - tree-sitter 0.20.2
  - tree-sitter-cpp 0.20.0
- tree-sitter-mozjs 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
  - tree-sitter-javascript 0.19.0
- tree-sitter-preproc 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-python 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-rust 0.19.0
  - cc 1.0.72
  - tree-sitter 0.19.3
- tree-sitter-typescript 0.20.0
  - cc 1.0.72
  - tree-sitter 0.20.2

As you can see tree-sitter-mozcpp and tree-sitter-typescript have been bumped to using tree-sitter 0.20.2 instead of 0.19.3. Interestingly, tree-sitter-cpp was already on 0.20.2 but tree-sitter-mozcpp was not (but now it is).

So my understanding (with very little Rust experience) is that the enum crate is using the higher tree-sitter and then any of the lower ones are returning the wrong type. In the example above it is java but I confirmed this by moving others up to the top of the match. If I try constrain tree-sitter to 0.19 then typescript and cpp return the wrong type.

I actually ran into this issue trying to consume ``rust-code-analysisin another crate and ended up searching for a version of the crate where thetree-sitter` versions aligned.

So @marco-c I am going to need some help on how to resolve this:

• Why did this work before? It seems to be purely due to when the lock file was generated? Or the lock file was edited manually? I confirmed this by deleting the lock file on ~HEAD and indeed, building throws the same error.
• What is the alternative here? With my limited knowledge I guess it would be keeping the tree-sitter versions in sync and only upgrading in lock-step?

I see there is a ticket for upgrading tree-sitter to 0.20 #674 so maybe a known thing that this should be in lockstep?

Sorry about the wall of text if this is all know stuff but it helped me attempt to reason through it.

[dburriss]

To try confirm the above I downgraded typescript and mozcpp and pinned the tree-sitter in enum to 0.19.3 and the enum project compiles again.

[marco-c]

I think we should try to have all grammars use the same tree-sitter version, otherwise we have too many complications. We should either update all to 0.20, or revert all to 0.19. CC @Luni-4