GeezerVison

…t that fits on one line

A trailing line comment lands in the level of the statement's last break, while
its forced line break lands in the enclosing level. When that level fits on one
line it is written from its flat text, and Comment.computeFlat added the space
after the slashes itself, bypassing the exemption that JavaCommentsHelper.rewrite
has for the IDE markers: `String a = "x"; //$NON-NLS-1$` came out as
`// $NON-NLS-1$`, which Eclipse no longer recognises.

The missing-space rule is now one method, JavaCommentsHelper.addMissingSpace,
applied by both paths, and the comment's width follows its flat text instead of
a hand-maintained + 1. A trailing `///` comment that already had its space is no
longer turned into `// /` either: 4 of the 15,747 JDK 21 sources change, all
that way.

The tests were written first: trailingLineCommentGetsItsMissingSpaceExceptIdeMarkers
and the ojf-issue-23-non-nls-trailing-marker fixture failed before the change.

The first of the three cases of #23. Same defect upstream:
google/google-java-format#221, google/google-java-format#361.

… statement is wrapped or reflowed

Eclipse numbers a marker by the position of its literal on the line and reads
it only there. The formatter left every marker of a wrapped call on the call's
last line, and gave a reflowed literal one marker, on its last piece.

OpsBuilder.build now pairs each marker of a markers-only trailing comment with
the n-th string literal of the comment's input line, within the same
partial-format region, and puts a zero-text ReservedWidth after that literal,
so the layout counts the marker on the literal's line and a second run changes
nothing. JavaOutput.appendNonNlsMarkers writes each paired marker at the end of
its literal's output line, numbered by the literal's position among the string
literals written there; a marker whose number has no literal stays where it
was. JavaCommentsHelper never wraps a comment of markers. StringWrapper
recomputes the markers of a reflowed chain's line from the new line structure,
one marker after every piece and the markers of the literals before and after
the chain renumbered for their lines, and reserves their room per line; a mixed
comment, a second changing chain on the line, or a chain that ends in a text
block leaves the markers alone.

The tests were written first: the fixtures ojf-issue-23-wrapped-arguments,
ojf-issue-23-reflowed-literal and ojf-issue-23-marked-literals, with
nonNlsMarkersStayOnTheLineOfTheirLiteral and aRunOfNonNlsMarkersIsNeverWrapped,
failed before the change. None of the 15,747 JDK 21 sources changes: the JDK
carries no Eclipse markers.

Fixes #23. Same defect upstream: google/google-java-format#221,
google/google-java-format#361.

The marker code used Guava's ImmutableList for its lists of marker numbers
and literal indices. List.of, List.copyOf and Stream.toList give unmodifiable
lists without it, so the signatures take and return List. The field in
Comment carries the Immutable suppression the other doc nodes use for their
effectively immutable state.

The fixtures exercise the marker code end to end only. NonNlsMarkersTest pins
which comments count as markers only, the order and the places in which the
numbers are read, and the spelling of a marker; ReservedWidthTest pins that a
reservation takes its width on the line and nothing else; the two
JavaCommentsHelperTest cases pin the missing-space rule and its exemptions.

google/MOE, the tool whose markers the line-comment wrapper kept on one line,
has been dead for years. Its special case goes, with the test that pinned it;
only the NON-NLS marker runs stay unwrapped.

The fork's own files carry the Apache 2.0 header without a copyright line.
The four files this change adds were created with a Palantir one; CLAUDE.md
records the rule, and #154 covers the rest of the tree.

Comment carried the annotation and needed a suppression for its List field,
and ReservedWidth had taken the annotation from its sibling nodes. Doc itself
is not annotated, so the check adds nothing here: the nodes are immutable by
construction.

The locals this change added declare their type with var where the
initializer gives it unchanged. The loop variables over a List<Integer> keep
int, so that they are not boxed, and the lambda keeps its target type.

CLAUDE.md: plain java.util collections instead of Guava's immutable ones,
javax.annotation.Nullable as the one nullability annotation, var for locals,
no Error Prone @immutable, and the Apache header without a copyright line on
a new file.