[circt-verilog-lsp] Move to Slang's src mngr, drop LLVM src mngr

[Scheremo]

• Use Slang buffers as source of truth

  • Replace llvm::SourceMgr-centric code with direct use of slang::SourceManager.
  • Add slang::BufferID mainBufferId and expose getMainBufferID().

• Interval map now indexes Slang buffer pointers

  • New typedef: using SlangBufferPointer = char const *;.
  • Change from IntervalMap<const char*, …> to IntervalMap<SlangBufferPointer, …> (half-open).
  • Lookups are computed from slang::SourceManager::getSourceText(buf).data() + offset.

• Remove LLVM SourceMgr plumbing

  • Drop members/APIs: sourceMgr, bufferIDMap, getSMLoc(...).
  • Update filePathMap to store (slang::BufferID, path) pairs.
  • getOrOpenFile(...) now assigns text via Slang (driver.sourceManager.assignText(...)) and caches the returned BufferID.

• Implement LSP UTF-16 to Slang UTF-8 conversion

  • Add decodeUtf8(...) and lspPositionToOffset(...) helpers.
  • Use these helpers in getLocationsOf(...) and findReferencesOf(...) before interval lookups.
  • Clamp positions past end-of-line to buffer end; avoid OOB pointer math.

• Definition & reference resolution updated

  • getLspLocation(slang::SourceLocation/Range) now uses Slang line/column and compares loc.buffer() directly to mainBufferId.
  • findReferencesOf(...) / getLocationsOf(...) locate the interval by pointer into the main buffer.

• Top module discovery via syntax trees

  • Derive driver.options.topModules from CompilationUnitSyntax members by reading ModuleDeclarationSyntax headers (covers modules and packages).
  • Avoid creating a separate Compilation just to enumerate tops.

[uenoku]

This looks fantastic, thank you for unifying them!

[Scheremo]

This looks fantastic, thank you for unifying them!

Thanks! I'll open this for review / merging now, do let me know if you have any points that stick out to you 😄
I'm happy with the current state, so feel free to merge if you don't find any issues.

[uenoku]

LGTM, thank you for improving buffer handling!

[Scheremo]

LGTM, thank you for improving buffer handling!

Thank you once more for the helpful and targetted feedback. You really helped to point out some areas to improve further!
I'm okay with merging at this point, but would be interested in your opinion on this:

I think unit tests make a lot of sense (easier to test things that way), but I would suggest to first break out VerilogServer, VerilogServerContext, VerilogTextFile and VerilogDocument into separate .cc files with corresponding header files. We could then add a unittest directory in the VerilogServerImpl directory, e.g. like this:

VerilogServerImpl/
├─ CMakeLists.txt
├─ include/
│ ├─ VerilogTextFile.h
│ ├─ VerilogDocument.h
│ ├─ VerilogServerContext.h
│ └─ VerilogServer.h
├─ src/
│ ├─ VerilogTextFile.cc
│ ├─ VerilogDocument.cc
│ ├─ VerilogServerContext.cc
│ └─ VerilogServer.cc
└─ unittests/
└─ VerilogDocumentTests.cpp

If something like this is fine for you, I can open a separate PR to implement it. Let me know what you think!

[uenoku]

I'm okay with merging at this point, but would be interested in your opinion on this:

Thanks! The current PR looks also great me!

with corresponding header files. We could then add a unittest directory in the VerilogServerImpl directory, e.g. like this:

I agree, I think this looks awesome! I also think separating headers is great. One nitpick is we probably want to use flat structures without include and src directories. LLVM somehow prefers this kind of flat directories for header files used under lib/ directories.

VerilogServerImpl/
|- CMakeLists.txt
|- VerilogTextFile.h
|- VerilogTextFile.cpp
|- VerilogDocument.h
|- VerilogDocument.cpp
...