Fix false positive grid_swizzle_conflict validation for Triton swizzled 1D grids

[mzweilin]

This PR replaces the validator’s line-based grid heuristic with an AST-based check so it only flags actual 2D grid launches when GROUP_SIZE_M swizzling is present. The previous implementation incorrectly treated a 1D single-tuple grid lambda as 2D whenever the line contained multiple triton.cdiv(...) calls.

It also adds regression coverage for both cases:

• a valid 1D swizzled grid that should pass
• a real 2D grid with swizzling that should still fail

Claude Code was smart enough to identify this false positive result, but it is better to fix the bug.

[Copilot]

Pull request overview

This pull request updates Xe-Forge’s Triton static validator to avoid false positives for grid_swizzle_conflict by replacing a line-based heuristic with an AST-based grid dimensionality check, and adds regression tests for 1D vs 2D swizzled grids.

Changes:

• Replaced the line-count/comma heuristic for detecting 2D grids (when swizzling is present) with an AST-based analysis of grid assignments.
• Added regression tests to ensure a 1D swizzled grid is allowed and a 2D swizzled grid is rejected.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/xe_forge/core/validator.py	Switches grid_swizzle_conflict detection from a string/line heuristic to AST-based detection for more accurate 1D vs 2D classification.
tests/test_validator.py	Adds tests covering allowed 1D swizzled grids and rejected 2D swizzled grids.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

[mzweilin]

@sandlbn ready to review

[sandlbn]

LGTM