Incorrect white-space auto-formatting inside Javadoc code tags

[yoshi-sys]

Consider the following code that auto-formats a Javadoc comment, represented as a String:

public static void main(String[] args) throws Exception {
  System.out.println(format("/**<pre><code>a\n b</code></pre>*/"));
  System.out.println(); System.out.println(); System.out.println();
  System.out.println(format("/**\n<pre><code>a\n b</code></pre>*/"));
}

private static String format(String code) throws Exception {
  var options = DefaultCodeFormatterOptions.getEclipseDefaultSettings().getMap();
  options.put(JavaCore.COMPILER_COMPLIANCE, JavaCore.VERSION_21);
  options.put(JavaCore.COMPILER_CODEGEN_TARGET_PLATFORM, JavaCore.VERSION_21);
  options.put(JavaCore.COMPILER_SOURCE, JavaCore.VERSION_21);
  var edit = ToolFactory.createCodeFormatter(options).format(CodeFormatter.K_JAVA_DOC, code, 0, code.length(), 0, "\n");
  var document = new Document(code);
  edit.apply(document);
  return document.get();
}

I would expect that the result of the two format calls be the same. However, the actual output is:

/**
 * <pre>
 * <code>a
b</code>
 * </pre>
 */



/**
 * <pre>
 * <code>a
 b</code>
 * </pre>
 */

The only difference between the first and the second output is that the first one does not contain a space before b, while the second one does.

It seems to me that the second output is the correct one, because I don't think text inside pre/code blocks should be formatted, and there is a space before the b in the source text. In any case, I don't see why adding a line break before the pre tag should change the output.

This was tested with version 3.39.0 of the Maven artifact org.eclipse.jdt:org.eclipse.jdt.core.

[yoshi-sys]

After some debugging, it looks like part of the issue is with the method TokenManager#getLength(int, int, int), which is called by findCommentLineIndent.

I do not fully understand why this is the case, but when handling the pre tag, for the first case, it calculates that length to be 0 because there is a new line right before the pre. However, for the second case, this length is calculated as 1, which I think, in turns, means that the single space before the b is considered "indent" and left out of the formatted text. In a debug session, if I force this length to be 0, then the 2nd output is printed for both cases.

This is tested with JDK 21.

[jukzi]

Please rewrite your testcase as Junit Test for JDT and provide a PR.

[yoshi-sys]

Hello,
I have added a JUnit test case demonstrating the incorrect behaviour with #3371