Fix class extraction for Rails' strict locals (tailwindlabs#19525)

Fixes: tailwindlabs#19481

This PR improves the Ruby extractor to better handle strict locals. We
recently introduced skipping comments in the Ruby extractor (PR tailwindlabs#19243
for tailwindlabs#19239) by ignoring comments that start with `#` until the end of
the line.

Strict locals are implemented like this:
```ruby
<%# locals: (css: "text-amber-600") %>
```

Notice the `#` after the `<%`, we considered this a comment and ignored
it.

This PR changes that behavior slightly where we skip comments that are
preceded by `%`. This means that `<%# anything here _will_ be scanned
%>`. This should solve the strict locals case, and normal comments will
still be skipped.

We can be more strict in the future if needed, but I think that this
should be a good solution for both scenarios.

### Test plan

1. Added a test to ensure we extract candidates in strict locals
2. Added a regression test for issue tailwindlabs#19239 where we introduced skipping
comments in the Ruby extractor
3. Other existing tests are still passing

We can also verify the extracted candidates:
(it's subtle, but you can see that the class is being extracted now)

<img width="1187" height="1376" alt="image"
src="https://github.com/user-attachments/assets/74bbfd79-9db4-4a5b-bd8d-25f1565c6bfd"
/>

Author: RobinMalfait

CHANGELOG.md

@@ -13,6 +13,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Allow whitespace around `@source inline()` argument ([#19461](https://github.com/tailwindlabs/tailwindcss/pull/19461))
 - CLI: Emit comment when source maps are saved to files ([#19447](https://github.com/tailwindlabs/tailwindcss/pull/19447))
 - Detect utilities when containing capital letters followed by numbers ([#19465](https://github.com/tailwindlabs/tailwindcss/pull/19465))
+- Fix class extraction for Rails' strict locals ([#19525](https://github.com/tailwindlabs/tailwindcss/pull/19525))
 
 ### Added
 

crates/oxide/src/extractor/pre_processors/pre_processor.rs

@@ -25,7 +25,33 @@ pub trait PreProcessor: Sized + Default {
     }
 
     #[cfg(test)]
-    fn test_extract_contains(input: &str, items: Vec<&str>) {
+    fn test_extract_exact(input: &str, expected: Vec<&str>) {
+        use crate::extractor::{Extracted, Extractor};
+
+        let input = input.as_bytes();
+
+        let processor = Self::default();
+        let transformed = processor.process(input);
+
+        let extracted = Extractor::new(&transformed).extract();
+
+        // Extract all candidates and css variables.
+        let candidates = extracted
+            .iter()
+            .filter_map(|x| match x {
+                Extracted::Candidate(bytes) => std::str::from_utf8(bytes).ok(),
+                Extracted::CssVariable(bytes) => std::str::from_utf8(bytes).ok(),
+            })
+            .collect::<Vec<_>>();
+
+        if candidates != expected {
+            dbg!(&candidates, &expected);
+            panic!("Extracted candidates do not match expected candidates");
+        }
+    }
+
+    #[cfg(test)]
+    fn test_extract_contains(input: &str, expected: Vec<&str>) {
         use crate::extractor::{Extracted, Extractor};
 
         let input = input.as_bytes();
@@ -46,7 +72,7 @@ pub trait PreProcessor: Sized + Default {
 
         // Ensure all items are present in the candidates.
         let mut missing = vec![];
-        for item in &items {
+        for item in &expected {
             if !candidates.contains(item) {
                 missing.push(item);
             }

crates/oxide/src/extractor/pre_processors/ruby.rs

@@ -119,7 +119,11 @@ impl PreProcessor for Ruby {
                 }
 
                 // Replace comments in Ruby files
-                b'#' => {
+                //
+                // Except for strict locals, these are defined in a `<%# locals: … %>`. Checking if
+                // the comment is preceded by a `%` should be enough without having to perform more
+                // parsing logic. Worst case we _do_ scan a few comments.
+                b'#' if !matches!(cursor.prev, b'%') => {
                     result[cursor.pos] = b' ';
                     cursor.advance();
 
@@ -382,4 +386,45 @@ mod tests {
         "#;
         Ruby::test_extract_contains(input, vec!["z-1", "z-2", "z-3"]);
     }
+
+    // https://github.com/tailwindlabs/tailwindcss/issues/19239
+    #[test]
+    fn test_skip_comments() {
+        let input = r#"
+          # From activerecord-8.1.1/lib/active_record/errors.rb:147
+          # Rails uses RDoc cross-reference syntax in inline documentation:
+          # {ActiveRecord::Base#save!}[rdoc-ref:Persistence#save!]
+        "#;
+
+        // Nothing should be extracted from comments, so expect an empty array.
+        Ruby::test_extract_exact(input, vec![]);
+    }
+
+    // https://github.com/tailwindlabs/tailwindcss/issues/19481
+    #[test]
+    fn test_strict_locals() {
+        // Strict locals are defined in a `<%# locals: … %>`, but the `#` looks like a comment
+        // which we should not ignore in this case.
+        let input = r#"
+          <%# locals: (css: "text-amber-600") %>
+          <% more_css = "text-sky-500" %>
+
+          <p class="text-green-500">
+            In a partial
+          </p>
+
+          <p class="<%= css %>">
+            In a partial using explicit local variables
+          </p>
+
+          <p class="<%= more_css %>">
+            In a partial using explicit local variables
+          </p>
+        "#;
+
+        Ruby::test_extract_contains(
+            input,
+            vec!["text-amber-600", "text-sky-500", "text-green-500"],
+        );
+    }
 }