Global variable not being correctly recognized

[Huang-Xuwei]

Hi, I’ve encountered an issue where the global variable c is not being correctly recognized in my project. The issue arises in the context of analyzing the symbol for c in the code, particularly when trying to use it within functions or across the document.

I’ve attempted to address this by modifying the TODO section in the ./LSPRAG/src/lsp/symbol.ts file. However, despite my changes, it seems that the recognition issue still persists, and c isn’t being detected as expected.

What I've done so far:

Modification: I updated the TODO section in the symbol.ts file, attempting to improve the handling of global variables.

Testing: After implementing the change, I tried rerunning the relevant functionality, but the issue remained unresolved.

Current Observation:

The variable c is still not correctly identified as a global variable and doesn’t seem to be properly handled during symbol analysis.

Request for guidance:
I would like to contribute to fixing this issue. Could you please guide me in the right direction? Specifically:

Are there any additional changes needed in the logic that handles global variables?

Should there be adjustments in other parts of the code (e.g., the LSP server or symbol processing logic)?

Any hints on what might be missing or incorrect in my current approach?

I’d really appreciate your advice on how to resolve this issue so that I can contribute a fix.

Thank you!

[Gwihwan-Go]

Thanks for reusing our LSPRAG project!

The issue you found mainly comes from two causes:

Problem 1:
LSPRAG currently extracts only tokens that are directly related to the control-flow graph (e.g., tokens appearing in if conditions or closely related statements).

Problem 2:
The current implementation does not properly handle constant variables.

---

To address this issue, both problems need to be handled.

The good news is that, in my opinion, these are not hard to fix 🙂
If you are interested in contributing to the project, we would be very happy to welcome your contribution, and I’m glad to share some conceptual guidance for the fix.

---

Suggested Fix Approach (Conceptual)

Switch to the active branch
Please check out the nightly branch first, as this is the actively maintained branch where new fixes are merged.

Add a new generation option
Since Problems 1 and 2 are expected behaviors of the current design, I recommend implementing your fix as a new option rather than modifying the default behavior.

Go to src/config.ts and define a new GenerationType.

Implement a new generator strategy

1. Create a new file under
   src/strategy/generators/{your_file_name}.ts
   Implement your fix there.
   Don’t forget to register the new strategy in factory.ts.

2. Fix for Problem 1 (CFG token restriction)

Skip the complex CFG-based extraction logic.
Directly use all tokens returned by:
await contextSelector.loadTokens();
Note: this may introduce more noise due to the larger token set, as discussed in our paper.

3. Fix for Problem 2 (constant variables)

Implement a dedicated context-loading strategy for symbols of type Constant.

A good starting point is to add logic similar to
src/agents/contextSelector.ts:addDefinitionToTerm.

Specifically:

(a) Inspect how the Language Server identifies global variables/constants. You can console.log the Symbol's type information. It is likely Constant or Variable, but needs confirmation.

(b) Implement a context-gathering strategy for global-variable–like symbols.
I recently found that using getHover is a robust way to retrieve constant-related information. You can likely reuse:

const hoverResults = await getHover(doc, symbol, false);
const hoverText = extractHoverText(hoverResults) || null;

---

Since the project is still under active development, some parts of the codebase may not yet be perfectly organized.
Feel free to ask any follow-up questions if you need help—I’m very open to discussion and patches!

Thanks again for your interest and contribution 🙌