Do not inline a lambda body whose closing paren carries a comment

[abashev]

Closes #62, which mirrors palantir#1792 (the reporter also points at palantir#1152).

The bug

A // comment on its own line before the )) that close a lambda's parenthesized body and the call around it was joined onto the code line before it, and the )) ended up inside the comment:

                        || items[i].getName().equals("gamma")// || (items[i].getName().equals("delta") && items.length >
                                                             // i// && items[i + 1].getName().equals("epsilon"))))

The output no longer parsed. The command line noticed only because its import pass parses the result again, and reported error: ')' expected at a position in the output; a caller of formatSource got the text as it was.

Why

An expression lambda's body is laid out by handle_breakOnlyIfInnerLevelsThenFitOnOneLine. When the body does not fit after -> but its first line does, tryInlinePrefixOntoCurrentLine lays the body's level out "on one line" through tryToLayOutLevelOnOneLine, which marks every break of that level as not taken and only lets the inner levels break. The comments before the closing ) belong to that same level, with forced breaks around them, and a forced break laid out flat is exactly what put the comment on the code line. The docs before the last inner level were checked for that (a forced break makes their width infinite); the docs after it were not checked at all. google-java-format has no such inlining, which is why it lays the input out correctly.

The fix

tryInlinePrefixOntoCurrentLine now refuses a level whose trailing docs have infinite width, the same check tryBreakInnerLevel already makes for its suffix. Such a body is not inlined and breaks normally: it starts on the line after ->, and the comments keep their own lines at the body's indent, which is what the formatter already did when the body's first line did not fit either:

                .filter(i ->
                        (items[i].getName().equals("alpha")
                                || items[i].getName().equals("beta")
                                || items[i].getName().equals("gamma")
                        // || (items[i].getName().equals("delta") && items.length > i
                        // && items[i + 1].getName().equals("epsilon"))
                        ))
                .findFirst()
                .orElse(-1);

A block comment in the same place went through the same path: it came out glued to the code without a space, bbb(i)/* block */));, and moved on a second run. It gets its own line now as well.

Checked

• The golden ojf-issue-62-comment-before-lambda-close-parens holds the reporter's input, the same shape as a plain call, and the block comment. It fails on main (--skip-* flags show the raw output) and passes with the change, and its output is stable on a second run.
• Related shapes format exactly as on main: a trailing // comment inside the body, a comment before the single ) of a call, a parenthesized argument that is not a lambda, a short body that already moved after ->, and a statement lambda with a comment before }.
• ./gradlew :open-java-format:test on JDK 21: 1541 tests, all green.
• The 15,747 files of the JDK 21 sources format exactly as before; none of them has a comment before )), and the five with a comment before a single ) do not change.

Version

Only input that failed today, or came out glued and unstable, formats differently, so this is a bug fix in the 2.x sense.