[clangd-dev] Implementing symbols in document (textDocument/documentSymbol)

Ilya Biryukov via clangd-dev clangd-dev at lists.llvm.org
Thu May 31 02:14:19 PDT 2018


Hi Marc-André,

I find that our index is just not suited for file-centric queries at this
point. E.g. it aggregates all redecls and chooses just one when creating
the symbol, has the logic to filter out non-local symbols, etc. Moreover,
it was design with project-wide queries in mind and changing it is
definitely a lot of work.

For file-major queries, we have the AST and I think we should use it
instead. So I suggest we:
1. Traverse the AST of the current file to find all decls inside this file.
The results will not contain everything, including local vars(do we need
them BTW?), static functions, etc. We can also convert results to something
more suitable for filtering, e.g. SymbolSlab. But I won't put them into the
index.
2. Search through results from step 1 and filter out the ones that don't
match the query.

If step 1 turns out to be slow, we can compute the results once and stash
them somewhere as an optimization. But given that the current file is
usually small enough, we can probably even get away without it.

--
Regards,
Ilya Biryukov


ср, 30 мая 2018 г. в 23:24, Marc-André Laperle via clangd-dev <
clangd-dev at lists.llvm.org>:

> Hi,
>
> I've been thinking about how to implement the "symbols in document"
> feature. I think it would make a lot of sense to reuse the in-memory index
> for this. One problem is that there is no current way to do "file-major"
> queries, i.e. get all symbols within one file.
>
> I see a few options:
> 1) Change the YAML format, in-memory model and index interface to allow
> file-major queries. This is the most work. On the other hand, I don't think
> it's worth investing a ton of effort in the YAML format.
> 2) Change only the "in-memory" index and interface for file-major queries.
> Not sure how feasible that is, but perhaps a good solution.
> 3) Run a new SymbolCollector every time on textDocument/documentSymbol,
> only on the needed file. That way there is a lot less symbol to filter out
> (only need the ones in the main file).
> 4) Iterate through all symbols, keep only the ones in the correct file.
> This is basically workspace/symbol with another "file" parameter. This is
> likely the least amount of work.
>
> I tried #4 and it seems to work fine but is probably not the most scalable
> so I am not sure how acceptable that is as an interim implementation.
>
> Let me know what you think,
> Marc-André
>
> _______________________________________________
> clangd-dev mailing list
> clangd-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/clangd-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/clangd-dev/attachments/20180531/5ca40725/attachment-0001.html>


More information about the clangd-dev mailing list