moonbitlang / core / 6483

89%

push

github

perf(json): fill indent cache incrementally, plus oracle test and common-path benchmarks (#4183)

* perf(json): build indent cache incrementally

Each cached indentation level is now derived from the previous one plus a
single indentation unit, instead of recomputing " ".repeat(indent * level).
This fills the cache in O(indent) per new level rather than O(indent * level),
and removes the `indent * level` product, which overflows Int to a negative
value for a large indent and is then rejected by String::repeat.

Adds an oracle-driven regression test covering indent widths 0..=6 crossed
with nesting depths 0..=12 (spanning both sides of the 8-space cutoff the
pre-cache implementation special-cased), plus compact-path and realistic
shallow-document benchmarks so the common paths are guarded.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014z81qJ5vdYoxZ48JTKCdAy

* refactor(json): use an array pattern for the indent cache tail

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014z81qJ5vdYoxZ48JTKCdAy

* fix(json): do not build the indent unit when no indentation is emitted

Codex review of #4183 caught a regression I introduced: hoisting the
indentation unit to one " ".repeat(indent) per stringify call made
stringify(indent=-1) abort, because String::repeat rejects negative counts,
where main took the compact path. It also allocated a huge unused unit for a
scalar document with a large indent.

The unit is now built inside write_indent, which callers reach only when
indent > 0, and only once per new cache level. Adds regression tests for
non-positive indent and for a huge indent on documents that emit none.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014z81qJ5vdYoxZ48JTKCdAy

* docs(json): correct the complexity claim on the indent cache fill

Codex revi... (continued)

3 of 3 new or added lines in 1 file covered. (100.0%)

16869 of 18537 relevant lines covered (91.0%)

283205.49 hits per line