Summary

codedb currently detects .tf, .tfvars, and .hcl files as unknown, so outline, symbol, and deps return nothing for Terraform/HCL codebases. Additionally, .terraform/ provider cache directories are not in skip_dirs, causing the index to balloon with thousands of cached module files.

Problem

On a real Terraform monorepo:

• 19,487 files indexed (17,000+ from .terraform/ caches)
• 18s snapshot time, 8.3MB snapshot
• 0 symbols extracted from .tf files (detected as unknown)
• Search results polluted by cached provider/module files

Proposed Solution

1. Add .terraform and .terragrunt-cache to skip_dirs in watcher.zig
2. Add hcl to the Language enum with detection for .tf, .tfvars, .hcl extensions
3. Implement parseHclLine() recognising top-level HCL block types:
   • resource "type" "name" → struct_def with symbol type.name
   • data "type" "name" → struct_def with symbol data.type.name
   • module "name" → import
   • variable "name" → variable
   • output "name" → constant
   • provider "name" → import
   • locals / terraform / moved → keyword blocks
4. Record source = "..." inside module blocks as imports for the dep graph
5. Add HCL comment support (#, //, /* */) to isCommentOrBlank

Results after fix

On the same Terraform monorepo:

• 1,836 files indexed (down from 19,487)
• 93ms snapshot time (down from 18s)
• Full symbol extraction across all HCL block types
• Clean search results with no .terraform/ noise

Files touched

• src/watcher.zig — 2 lines (add to skip_dirs)
• src/explore.zig — ~240 lines (language enum, detection, parser, helpers)
• src/tests.zig — ~250 lines (11 test cases)

Total: 493 inserted lines.

Tests

All existing tests pass (zig build test exit 0). 11 new tests cover:

• Each HCL block type (resource, data, variable, output, module, provider, locals, terraform)
• Full multi-block file parsing
• Comment skipping
• findSymbol integration
• .tfvars language detection
• isCommentOrBlank for HCL