Stop splitting a chained invocation that would fit on one line

[ctgnz]

Closes #152.

// before                                          // after
this.entityTypes.addAll(entity.getEntityTypes()    this.entityTypes.addAll(entity.getEntityTypes().stream().map(this::adaptEntityType).toList());
    .stream()
    .map(this::adaptEntityType)
    .toList());

Two settings, and neither works alone

alignment_for_selector_in_method_invocation = 85
    = 80 (M_NEXT_PER_LINE_SPLIT) + 4 (M_INDENT_BY_ONE) + 1 (M_FORCE)

M_FORCE splits regardless of fit, so it drops to 84. On its own that changes nothing, because join_wrapped_lines was false and the Eclipse formatter then only ever adds a wrap, never removes one already present — so that moves to true as well. lineSplit stays at 200.

That interaction is the whole reason this looked like a formatter-choice problem rather than a two-line config problem.

Measured

files reformatted	175
net change	1,316 insertions, 2,676 deletions — ~1,360 lines shorter
clean verify	green on -Pstandard, -Phistorical, -Pbattleorder
generator templates	unaffected — raw regeneration then one spotless:apply lands exactly on the committed sources

That last row was the thing worth checking, and it is why this is cheap: the zero-diff property CLAUDE.md describes still holds, so no template needed retuning.

Quality rather than just height — of 147 newly joined lines in hand-written code, 138 are ≤90 characters and the longest is 114, against a limit of 200:

Files.walk(packageRoot).map(Path::toFile).sorted(Comparator.reverseOrder()).forEach(File::delete);
library.getHqtfDummys().stream().filter(status -> status.isSupported(symbolSet)).forEach(hqtfDummy -> {

The 21,000-character lines in the generated libraries are injected SVG string literals, byte-identical before and after. A string constant cannot be wrapped, and this does not make them worse.

It stays the Eclipse formatter, deliberately

Re-import the profile and the IDE's Source → Format agrees with the build again. A switch to palantir-java-format or google-java-format would have given that up, reformatted every generated file, and needed the templates retuned — for a chain style this already achieves.

The trade-off

join_wrapped_lines = true means the formatter now rejoins lines that were wrapped on purpose, which is presumably why it was false. Where a hand layout genuinely matters, // @formatter:off / // @formatter:on still work.

Worth reviewing the diff for any place where a deliberate layout was worth keeping — it is mechanical, so anything that reads worse is a candidate for a @formatter:off rather than a reason to drop the change.

🤖 Generated with Claude Code