fix(twig): count {# #} comments - #798

What's broken

Twig writes a comment as {# ... #} (Templates, Comments), and the default tag_comment in Twig's src/Lexer.php is ['{#', '#}']. A comment runs from {# to the first #} after it, across lines, and comments do not nest.

The Twig Template entry in languages.json (line 8234) declares no comment at all:

"line_comment": [],
"multi_line": [],

so scc never reports a comment in a Twig file. Every line of a {# #} comment counts as code, including the docblock that opens 495 of Drupal core's 707 templates, and a tag commented out inside {# #} still adds to complexity.

Repro

{#
/**
 * @file
 * Displays the title of a page.
 */
#}
{# The heading is left out when there is no title. #}
{% if title %}
  <h1>{{ title }}</h1>
{% endif %}
{# {% if subtitle %}<h2>{{ subtitle }}</h2>{% endif %} #}

Across the 707 .twig files in Drupal core (cf863a1), 10,231 of the 22,518 lines move from code to comments, and code drops from 22,061 to 11,836.

The fix

"multi_line": [["{#", "#}"]] for Twig Template, the pair the Jinja entry already uses, plus the matching entry in processor/constants.go. The #} closer also ends the -#} and ~#} whitespace control forms, and leaving nestedmultiline off matches Twig. LANGUAGES.md lists names and extensions only, so it does not change.

One limit: scc can't tell {% verbatim %} text or a string literal inside a tag apart, so a {# in one of those (a Handlebars {{#each}} kept verbatim, say) opens a comment. Everywhere else Twig's own lexer reads {# as a comment opener too, so the two agree.

Verification

processor/workers_twig_comment_test.go adds four CountStats tests: comments on their own line (including the {#- -#} and {#~ ~#} forms), a multi-line docblock, a commented-out {% if %} that no longer adds complexity, and a guard where a comment shares a line with markup and the line stays code. The guard passes on master too.

There is no Go toolchain on the machine I made this on, so I haven't run go test or go generate myself. CI will. I worked out the expected counts with a line-for-line Python port of countLoopGeneric and the state functions it calls. That port reproduces the counts asserted by 69 existing CountStats tests in processor/. By the port, the three behaviour tests fail on master (Expected 4 code got 7, Expected 3 code got 12, Expected 1 code got 6) and all four pass on this branch. The Drupal figures above come from the same port. The processor/constants.go hunk is byte for byte what a Python port of scripts/include.go produces, and that port reproduces master's constants.go exactly.

I explicitly licence this contribution under the MIT licence.

🤖 Generated with Claude Code