[jonasporto]

Touches marcoroth/reactionview#103.

# herb:debug disable is only read on a block's opening tag (include_debug_disable_comment? runs against @erb_block_stack in in_excluded_context?), so a single output tag has no way to opt out of the debug span. Where that span lands in a place the HTML parser does not allow a <span>, the element it wraps is dropped or moved:

• A partial whose root element comes out of a helper, sent in a Turbo Stream. <%= tag.td "Ada", id: "cell_1" %> arrives as <template><span style="display: contents;"><td id="cell_1">Ada</td></span></template>; inside <template> the parser switches to "in body" at the <span> and ignores the <td> start tag, so only the text is left and the next stream that targets cell_1 finds nothing. <tr> and the other table-structure tags behave the same (reproduction in the issue).
• A layout whose <head> content comes from a partial. The partial's own AST has no <head> ancestor, so <%= stylesheet_link_tag … %> is wrapped and the browser closes <head> at the <span> (the issue's report).

With this change the comment is honoured on an output tag too, with the same whitespace rules as on a block:

<%= tag.td "Ada", id: "cell_1" # herb:debug disable %>

compiles to _buf << (tag.td "Ada", id: "cell_1").to_s; with no span, and the other output tags in the same template keep theirs. The engine already drops a trailing comment from the compiled Ruby, so nothing changes at runtime.

This is an opt-out, not a detection: at compile time the visitor cannot know that a partial will be sent inside a <template> or rendered into <head>, and the span is only wrong there. If you would rather skip the span automatically for the helpers whose output is a table-structure element (tag.td, tag.tr, content_tag(:td, …), the list in the HTML table model), that fits beside complex_rails_helper? and I can add it. The head case has no finite helper list (the report's template uses vite_stylesheet_tag), so the comment seems the right tool there.

Tests: three in test/engine/debug_mode_test.rb with snapshots, placed after the last top-level test so the existing snapshot numbers do not move. Against main they fail with the span around the helper's output. Checked against v0.11.1 and main at 5d75561 (debug_visitor.rb is identical in both). bin/integration locally (macOS, Ruby 4.0.7): 5256 Ruby tests (0 failures, 0 errors, 59 skips), 135 C checks, 0 leaks, bin/lint, bin/tidy, RuboCop with no offenses.