wz1000

On a project with 3000 modules, it looks like hls-graph spawns and hold on to 31000 TSOs for the duration of the entire build.

Yes, .hie files contain all the resolved names in a file, and can be generated by ghc 8.8+. See https://gitlab.haskell.org/ghc/ghc/wikis/hie-files and https://github.com/wz1000/HieDb for an example of how to consume the...

The parser is here, but I think it could be massively simplified and brought in line with all the other queries: https://github.com/wz1000/HieDb/blob/master/src/HieDb/Types.hs#L264

I think we can just accept the format used by all the other commands. These commands also have graceful fallback in case information like the module name, package name isn't...

I agree that we can improve the help message. However, the Symbol type can also be changed to make the module and unit-id optional, with the appropriate modifications to the...

Alternatively, we could consolidate all the input types into a single, easy to specify format like `unit-id:Module.symbol`, where both the unit-id and Module are optional.

Ah, in that case, it might be best to just document the existing format and leave everything else as is for now.

Sounds reasonable. But does this actually manifest in practice? Don't clients make another query when you refine your search by adding more characters?

I don't think it is possible for a sqlite database to be written to concurrently, even with the WAL and WAL2 modes. In particular, the COMMIT steps require exclusive access...

I'm not sure `hiedb` is the right place for this functionality, because different clients might expect different behaviour from `hiedb` depending on their own needs and usage pattern, like whether...