[yzxcj797]

Summary

.liquid files were not recognized by detect_language, so full builds excluded them and file summaries returned zero nodes.

This adds a structural, regex-based Liquid parser that indexes:

• a File node for every template;
• Variable nodes for assign declarations, including whitespace liquid blocks;
• Output nodes for non-empty {{ ... }} expressions;
• REFERENCES edges for static render and include targets.

Liquid comments are masked while preserving byte offsets and line numbers, so directives inside comments are not indexed.

Fixes #348.

Testing

• Added regressions for:
  • ordinary assign, output, render, and comment masking;
  • whitespace {% liquid ... %} blocks.
• pytest tests/test_liquid.py -q — 2 passed
• ruff check code_review_graph/parser.py tests/test_liquid.py — passes
• git diff --check — passes

[github-actions]

code-review-graph review

Overall risk: 0.40 (MEDIUM) — 8 changed function(s)/class(es), 17 affected flow(s), 5 test gap(s)

Risk-scored changes

Risk	Level	Symbol	Location	Tested
0.40	medium	code_review_graph/parser.py::CodeParser.line_at	code_review_graph/parser.py:3612	no
0.35	low	code_review_graph/parser.py::CodeParser._mask_liquid_comments	code_review_graph/parser.py:3580	no
0.30	low	code_review_graph/parser.py::CodeParser	code_review_graph/parser.py:2936	yes
0.30	low	code_review_graph/parser.py::CodeParser.replace_comment	code_review_graph/parser.py:3582	no
0.25	low	tests/test_liquid.py::test_liquid_file_has_file_assignment_and_render_nodes	tests/test_liquid.py:8	(test)
0.10	low	code_review_graph/parser.py::CodeParser._extract_bytes	code_review_graph/parser.py:3200	no
0.10	low	code_review_graph/parser.py::CodeParser._parse_liquid	code_review_graph/parser.py:3590	no
0.05	low	tests/test_liquid.py::test_whitespace_liquid_tag_is_indexed	tests/test_liquid.py:44	(test)

Affected execution flows

• visit_FunctionDef — criticality 0.41, 55 node(s) across 1 file(s)
• visit_AsyncFunctionDef — criticality 0.41, 55 node(s) across 1 file(s)
• visit_ClassDef — criticality 0.41, 54 node(s) across 1 file(s)
• visit_If — criticality 0.36, 3 node(s) across 1 file(s)
• repo_relative_path — criticality 0.36, 9 node(s) across 1 file(s)
• ...and 12 more affected flow(s)

Test gaps

• code_review_graph/parser.py::CodeParser._extract_bytes (code_review_graph/parser.py:3200)
• code_review_graph/parser.py::CodeParser.replace_comment (code_review_graph/parser.py:3582)
• code_review_graph/parser.py::CodeParser._mask_liquid_comments (code_review_graph/parser.py:3580)
• code_review_graph/parser.py::CodeParser._parse_liquid (code_review_graph/parser.py:3590)
• code_review_graph/parser.py::CodeParser.line_at (code_review_graph/parser.py:3612)

Token savings: this graph-backed report used ~174,355 fewer tokens (~89%) than reading every changed file in full (estimated, chars/4 approximation).

---

Powered by code-review-graph — local-first analysis; no code leaves the CI runner.

[tirth8205]

Outside this sweep's mission of demonstrated coding-session token savings: this adds a new Liquid parser. Bare template references, raw-block handling and a complete cross-file impact workflow need evidence before that language expansion is included.

[tirth8205]

Checks pass and the code looks right. Missing: tests for the edge cases it introduces.

• Test that {% raw %}...{% endraw %} content is not indexed (currently produces false-positive Output/Variable/REFERENCES nodes; the fix would mask raw blocks the same way comments are masked).
• Test that '#' comment lines inside {% liquid %} blocks are not indexed (currently 'assign'/'render' on '#' lines produce false positives).
• Test for {% include 'x' %} and for double-quoted render/include targets (include is never exercised).
• Test for empty file and for empty {{ }} output being skipped.
• Test for trim markers ({%- -%}, {{- -}}) and multi-line {{ ... }} output line ranges.

Every new branch in the code should have a test, including the failure branch. Disagree with any of these? Say so here first.

PRs now target staging, not main. Yours was retargeted already, so nothing to do there.

[tirth8205]

{% raw %} blocks are indexed as live code, and so are # comment lines inside a liquid tag. Both reproduce on the head commit, and both are among the five tests I asked for on 15 September. Those tests are not in the branch and there was no reply here. The only commit since is the staging merge on 18 September, and tests/test_liquid.py still has exactly two tests.

A file containing only a raw block:

{% raw %}
{% assign should_not_exist = 1 %}
{{ should_not_be_output }}
{% render 'not-a-real-snippet' %}
{% endraw %}

yields a Variable should_not_exist, an Output output_3:should_not_be_output, and a REFERENCES edge to not-a-real-snippet. None of it should reach the graph. _mask_liquid_comments masks {% comment %} and nothing else.

Same for # lines inside a liquid tag:

{% liquid
  # assign ghost_var = 1
  # render 'ghost-snippet'
  assign real_var = 2 %}

Variables come out as ['ghost_var', 'real_var'] with a phantom REFERENCES edge to ghost-snippet.

Render targets still do not resolve. I built Shopify Dawn on this branch. All 127 new liquid REFERENCES edges are stored with target_resolution='unresolved' against a bare string such as card-product or cart-drawer. snippets/card-product.liquid is a File node in the same graph with 0 inbound edges, even though 7 templates render it. "Who renders this snippet" returns nothing, so the cross-file impact this is for is not there yet.

Output names embed the line number, so qualified_name is <file>::output_{line}:<expr>. I inserted one {% comment %} line at the top of sections/image-banner.liquid and ran code-review-graph update. 30 of that file's 44 nodes were destroyed and recreated under new names. No other node kind behaves this way, and memory.py, graph_diff snapshots and the embeddings index all key on qualified_name. The same build emits 4046 liquid nodes over Dawn but stores only 3836. 210 collapse across 156 names through UNIQUE(qualified_name), for example card-product.liquid::ratio six times, so line_start lands on an arbitrary occurrence.

Gates are genuinely clean: 4237 passed, 0 failed, ruff clean, mypy clean. The table diff over Dawn is purely additive. 0 node names lost, 0 edges lost, Class 51 to 51, Function 429 to 429, CALLS 2522 to 2522. Nothing existing is touched or respelled.

One cost to weigh while fixing the rest. 3445 of 4352 nodes are Output nodes whose name is raw expression text. FTS hits for product go from 67 to 1151, and 85% of those are Output nodes. That works against the token-efficiency goal on exactly the repos this targets.

To merge:

• Mask {% raw %} blocks and # lines the way comments are masked, each with a test.
• Resolve render and include targets to the snippets/<name>.liquid File node so inbound edges exist.
• Drop the line number from Output names and make the key unique without it.
• Add the remaining three tests from my earlier list, plus tests/fixtures/sample.liquid.

[tirth8205]

Thank you for this PR.

I tested it merged with the latest staging on a small Liquid repo.

Before I can merge it, please fix these:

1. {% raw %} blocks are read as live code. So are # lines inside {% liquid %}. The parser skips only {% comment %} (code_review_graph/parser.py:3580). Example: {% raw %}{% assign ghost = 1 %}{% endraw %} gives a Variable ghost. Please skip both, each with a test.
2. render and include targets are saved as bare names like card (parser.py:3652). So snippets/card.liquid gets no inbound edges. Its impact radius is empty. Please link each target to its snippets/<name>.liquid File node.
3. Output names hold the line number (parser.py:3622). I added one line at the top of a file. Every Output node below it got a new name. Please make names unique without the line number.
4. Two assign ratio lines in one file become one node. The first one is lost (parser.py:3640).
5. Some tests are missing. Please test include, double quotes, an empty file and empty {{ }}. Please also test trim markers and multi-line output. Please add tests/fixtures/sample.liquid too.

After you push the fix, I will test it again.

[Copilot]

🟡 Changes recommended

The parser still extracts false variables, outputs, and dependencies from quoted values, inline comments, and raw blocks, polluting the code-review graph.

3 open issues

• Mask raw blocks during dependency extraction · New
• Exclude inline and Liquid-block comments from extraction · New
• Ignore quoted text when extracting Liquid directives · New

What changed in this PR

Adds structural Liquid template indexing to the code-review graph, addressing #348 by recognizing .liquid files and extracting graph content without a dedicated Tree-sitter grammar.

Changes:

• Registers Liquid file detection and a regex-based parser.
• Extracts file, assignment, and output nodes plus static render/include references, while masking block comments.
• Adds regression tests for ordinary tags and multiline liquid blocks.

File	Description
tests/​test_liquid.py	Tests Liquid detection, extraction, block-comment masking, and multiline blocks.
code_review_graph/​parser.py	Adds Liquid recognition, parser dispatch, and structural extraction.

_{Review effort: Balanced}