REWRITE ALL THE THINGS

[regexident]

I started refactoring the project, moving things from depending on 'rustc_…' (rustc internals) to 'ra_ap_…' (rust-analyzer internals)

Progress:

• $ cargo modules tree (now called $ cargo modules generate tree)
• $ cargo modules graph (now called $ cargo modules generate graph)

Since we're moving from performing analysis on a project's AST to analysis on its HIR we should be able to fix a bunch more issues in the long term, which right now see kind of hard if not impossible to fix:

• Edition 2018 Support
• Doesn't understand the crate keyword
• Doesn't understand cfg-if
• Differentiate between modules with the same name
• Support for cargo workspaces

It should also make these features feasible:

• Feature request: Option to list all public types
• Identify cycles between imports
• Analyze crates in non-test mode

---

(Once finished this will fix #63.)

[regexident]

Orphans working again in cargo modules tree

[kvark]

Please let me know when it would make sense to start testing it! Super excited to get that ball rolling on WR, once available.
Today, running the tree command isn't producing anything for me, just:

compositor_windows: pub

[regexident]

@kvark: right now the target selection does a very, very hack-ish "select first library found", iirc.

As such it could be possible that the logic somehow picked the compositor-windows package as target. The compositor-windows crate contains just a single modules, which matches your terminal output.

I still need to add a proper CLI target selection feature, as well as add proper support for workspace handling.

Target selection is the next thing I'll be working on, as for the cargo modules graph logic I'll need to do some more reading up on how rust-analyzer handles imports (I also think the graph should not be restricted to explicit imports, but any used paths per module).

---

If your wanted to test the new output of cargo modules tree today I'd suggest cloning the rust-lang/cargo project and testing it via cargo run --release -- tree /path/to/cargo. It's one of the projects I'm using to test things.

Quick heads-up: You will notice the tool is excruciatingly slow compared to current master (even when running as release build). This is due to the switch from analyzing rustc's pure AST to analyzing rust-analyzer's HIR, which is a more involved process. There's no free lunch, sadly.

[kvark]

Ok, I see. Thank you for the info! Graph is the stuff I need, and I'm waiting eagerly for it to be available :)

[regexident]

@kvark Commit f7d226f removed any remaining ra_ap_hir values from the graph's nodes, which were preventing persisting the graph to disk.

As such adding a --cache <FILE> (and a complementary --ignore-cache, I suppose) should now just be a matter of some minor elbow grease.

The idea would be to have cargo-modules skip the src -> AST -> HIR -> graph steps and jump straight to the graph, loading it from disk if a previously persisted graph is found at <FILE>. The persisted graph would be tagged with cargo-module's current version and outdated caches ignored.

[kvark]

So that would allow me to re-launch it with different focus points quickly? Or is it something else?

[regexident]

Once implemented, yes. That's the idea at least.

But first I want to update the readme to reflect the changes, maybe add a CHANGES file, merge this PR and publish a v0.5.

Just wanted to update you on the progress.

[pksunkara]

It looks like you implemented the hir visitor, but I have been trying to find that in this code (so that I can extract it as we talked about) and gave up lol.

[regexident]

@pksunkara I did, but it ended up being cleaner for my use-case to just explicitly drive the module walking via recursion, so I removed the visitor in d9ed9ae. 😄

I do still plan however to create a PR with it for cargo-up, as it should be quite useful for others. I just haven't gotten to it yet. 😇

[pksunkara]

🙏

I am just happy that someone else is finally using the ra_ap_* crates. I put quite a bit of work into getting that release process done.