[PATCH] D51628: [clangd] Implement a Proof-of-Concept tool for symbol index exploration

[Kirill Bobyrev via Phabricator via cfe-commits]

kbobyrev added inline comments.


================
Comment at: clang-tools-extra/clangd/dexplorer/Dexplorer.cpp:39
+
+// FIXME(kbobyrev): Make this an actual REPL: probably use LLVM Command Line
+// library for parsing flags and arguments.
----------------
ilya-biryukov wrote:
> kbobyrev wrote:
> > ilya-biryukov wrote:
> > > Maybe we could expose `CommandLineParser` from `llvm/lib/Support/CommandLine.cpp` as a public API and use it here?
> > > Not sure if there are any obstacles to doing so or how much work is it, though.
> > > E.g. `cl::opt` seem to rely on being registered in the global parser and I'm not sure if there's an easy way out of it.
> > Yes, that might be tricky. I have thought about a few options, e.g. consuming the first token from the input string as the command and dispatching arguments parsed via `CommandLineParser` manually (discarding those which are not accepted by provided command, etc). There are still few complications: help wouldn't be separate and easily accessible (unless hardcoded, which is suboptimal).
> > 
> > Another approach I could think of would be simplifying the interface of each command so that it would be easily parsed:
> > 
> > * `> fuzzy-find request.json`: this would require `FuzzyFindRequest` JSON (de)serialization implementation, but that would be useful for the benchmark tool, too
> > * `> lookup-id symbolid`
> > * `> load symbol.yaml`/`swap symbol.yaml`
> > * `> density-hist`
> > * `> tokens-distribution`
> > * `> dense-tokens`
> > * `> sparse-tokens`
> > 
> > And also a generic `> help` for the short reference of each command. Out of all these commands, only Fuzzy Find request would benefit a lot from a proper parsing of arguments, having JSON file representation might not be ideal, but it looks OK for the first iteration for me. Fuzzy Find request would presumably be one of the most used commands, though, but then again, we could iterate if that is not really convinient.
> The single-argument and no-arg commands certainly look nice. However, creating a separate file for each fuzzy-find request feels like a pain.
> 
> Have you tried prototyping parsing with `CommandLineParser `? What are the complications to exposing it from LLVM? 
I didn't, but I looked closely at `clang-query` which decided not to do that IIUC due to the complications with the `CommandLineParser` and also the specifics of the tool and `clang-refactor` which utilizes `CommandLineRefactoringOptionVisitor` and `RefactoringActionSubcommand` internal implementations, but overall I don't think it looks very simple.

We had a discussion with Eric and Sam, the consensus was that we should start with something very simple and iterate on that if we decide that more complicated interface is necessary.


https://reviews.llvm.org/D51628