v2.6.0-alpha

Full Changelog: v2.5.1-alpha...v2.6.0-alpha

v2.5.1-alpha

Full Changelog: v2.4.1-alpha...v2.5.1-alpha

v2.4.1-alpha

Full Changelog: v2.3.0-alpha...v2.4.1-alpha

v2.3.0-alpha

What's Changed

• deps: bump golang.org/x/net from 0.55.0 to 0.56.0 in the go-dependencies group across 1 directory by @dependabot[bot] in #33

New Contributors

• @dependabot[bot] made their first contribution in #33

Full Changelog: v2.2.0-alpha...v2.3.0-alpha

v2.2.0-alpha

What's Changed

• Deep-review fixes: full pass — parser/clipboard/release + 9 correctness & hardening fixes (#94–#105) by @bpacholek in #107

Full Changelog: v2.1.7-alpha...v2.2.0-alpha

v2.1.7-alpha — markup raw-text fidelity + parse hardening

Fixes from the pre-launch production-readiness review. The exported /v2 surface is unchanged; stays -alpha.

Fixed

• HTML <script>/<style> bodies and XML <![CDATA[…]]> are preserved byte-for-byte on Reformat instead of being whitespace-collapsed and entity-escaped. Escaping a script body's < to < produced invalid JavaScript (and a // line comment could collapse onto the next statement); these raw-text spans now round-trip verbatim. Ordinary element text is still escaped as before (#85).
• A panic in the raw fallback or the model finish step can no longer escape Parse. The recover boundary now covers the whole function — not just the structured parsers — degrading to the raw fallback (or an empty document) rather than crashing the embedding host (#86).

Documentation

• README documents the raw-text fidelity and the ordinary-text whitespace canonicalization (#85); fixed a misattached TypedKey/KeyDown doc comment and a stale "JSONC is not rewritten" note (#89).

Remaining production-review items (#87 #88 #90 #91 #92) are tracked in the v2.2.0 — production hardening milestone.

Full Changelog: v2.1.6-alpha...v2.1.7-alpha

v2.1.6-alpha — zero-allocation live reproject

The first half of #84 (the #80 follow-up): the live edit reproject now reuses a single segment-build scratch across all physical lines, instead of allocating one temporary per line.

Performance

• Combined with #80's arena pooling, a steady-state keystroke reproject of a large (>2 MiB) buffer now allocates 0 B / 0 allocations on both minified and pretty-printed shapes — the pretty shape previously allocated ~2.16 MB / ~67 000 times (one []Seg per line). The non-pooled ParseEditableColored benefits too (~67 000 → ~18 allocations). Output is byte-identical, locked by the TestEditPoolEqualsFresh equality fuzz and a new TestEditPoolPrettyAllocsConstant teeth guard.

The remaining half of #84 — making per-keystroke wall time strictly proportional to the edited region (the in-place line-range splice that revisits the immutable-Document decision) — is still open and P3.

The /v2 surface is unchanged; stays -alpha.

Full Changelog: v2.1.5-alpha...v2.1.6-alpha

v2.1.5-alpha — production-hardening follow-ups

Production-hardening follow-ups — the remaining issues from the code- & test-quality reviews (milestone v2.2.0 — production hardening). The exported /v2 surface is unchanged (TestExportedSurfaceGolden frozen); this stays -alpha (pre-production maturity, not API churn).

Fixed

• XML/HTML Reformat no longer decodes entities into the buffer — it re-encodes &→&, <→< (and "→" inside attribute values), so a reformat-then-save stays valid markup and round-trips & (#81). Display/copy still show the decoded text.
• A JSONC document that begins with a comment auto-detects as JSONC (a // license header or { // note) instead of falling back to raw, preserving the comment as a visible node; a // inside a string value still stays plain JSON (#82).

Performance

• The live edit reproject pools its Document arenas + buffer snapshot across keystrokes (#80): a keystroke on a large over-budget buffer drops from ~18.6 MB / 18 allocs to ~32 B / 1 alloc and is ~35 % faster, byte-identical to a fresh parse (locked by a randomized equality fuzz). Wall-time stays O(buffer); work strictly proportional to the edited region is deferred to #84.

Changed

• Dropped the dead tabWidth parameter from the internal ParseEditableColored (#83).

Docs

• Reconciled the contradictory -alpha / "v1.0.0 frozen" / "pre-1.0 v0.x" copy to one story across README / package docs / SECURITY.md / CHANGELOG, and added a testable drop-alpha checklist to WORKFLOWS.md (#67).

Full changelog: see CHANGELOG.md.

Full Changelog: v2.1.4-alpha...v2.1.5-alpha

v2.1.4-alpha

Third quality review — fixes + mutation testing (no public API change)

Performance

• Reflow no longer re-allocates the wrapped line's break list every scroll tick. It was a fresh []int32 local while the highlight passes already pool the identical scratch — ~895 KB of garbage per reflow on a 2 MB single wrapped line, on the Fyne UI goroutine. Now reuses a persistent r.reflowBreaks. Guarded by TestReflowReusesWrapBreaks.

Tests & tooling

• make mutation — mutation testing via gremlins over the pure logic packages, gating on test efficacy. Empirically, internal/geometry, internal/model, and internal/parse all score 100% test efficacy (0 surviving mutants) — the tests detect behavioral changes, not just cover lines.
• Gutter digit-width boundary-crossing guard (closes a teeth gap in the #77 memo).
• Folds in the Windows CRLF golden CI fix (CI is green on Linux/macOS/Windows).

Review verdict

No high-severity findings; the hardening-era tests are genuinely teeth-bearing. The "used-only-by-tests" helpers (maxTextWidth/LineString/ProjectionBytes) are intentional memory-invariant instrumentation. Follow-up filed: #83 (drop the dead tabWidth param).

Full diff: v2.1.3-alpha...v2.1.4-alpha

Full Changelog: v2.1.3-alpha...v2.1.4-alpha

v2.1.3-alpha

Production hardening (P2) — no public API change

Robustness (#76)

• keyExtend guards a nil renderer before reading r.firstRow (a pre-render call no longer panics).
• scanString clamps on a trailing backslash so an unterminated "\ can't push a node span past EOF.
• recomputeMetrics only locks the cell memo when it measured a non-degenerate cell — a transiently-unavailable font can't permanently render a 1px grid.
• Documented that live parse validity isn't refreshed above the MaxEditBytes cap (the indicator freezes rather than going stale-wrong).

Performance (#77)

• serializePretty presizes the spans slice; remapCaretOffset binary-searches (locked against the old linear logic over 3000 random cases).
• applyGutter memoizes the digit count on the line count.
• Regression benches added for big-document plain/worst-case-regex search and wrap reprojection.

Tests & deps

• (#78) Three self-erasing t.Skips → t.Fatal; a deterministic guard that valid structured input never silently regresses to the raw fallback (surfaced #82: JSONC starting with a comment auto-detects as raw).
• (#79) go.mod lowered from the patch-pinned go 1.26.4 to go 1.25 (the real floor, set by x/net/x/image) + toolchain go1.26.4; govulncheck unpinned to @latest; dependencies documented; a written deprecation policy added.

Follow-up filed: #82 (JSONC leading-comment auto-detect).

Full diff: v2.1.2-alpha...v2.1.3-alpha

Full Changelog: v2.1.2-alpha...v2.1.3-alpha