Skip to content

WO19 G6: pin the capital-U escape rows; document the rs grammar extensions - #279

Merged
ronaldtse merged 1 commit into
mainfrom
wo19-u-escape-rows
Nov 14, 2026
Merged

ronaldtse merged 1 commit into
mainfrom
wo19-u-escape-rows

Conversation

@ronaldtse

@ronaldtse ronaldtse commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Item 4 — \U escape pins. Embedded test rows for the six bel/rus maps that carry capital-\U escapes:

map row spelling pin
gki-bel-Cyrl-Latn-2000 "\U040E" (true escape) "Ў" -> "Ŭ"
mvd-bel-Cyrl-Latn-2008 "\U040E" (true escape) "Ў" -> "W"
odni-bel-Cyrl-Latn-2015 "\U040E" (true escape) "Ў" -> "W"
alalc-bel-Cyrl-Latn-1997 "\\U040E" (doubled backslash) "Ў" -> "Ў" (dormancy)
bgnpcgn-bel-Cyrl-Latn-1979 "\\U040E" (doubled backslash) "Ў" -> "Ў" (dormancy)
mvd-rus-Cyrl-Latn-2008 "\\U040E" (doubled backslash) "Ў" -> "Ў" (dormancy)

The three true-escape rows are the ones the generated ts/py parsers used to get wrong (from-value carried as the literal text "U040E"; the row never fired) — the pins make the embedded-test gates catch that regression in every runtime. Verified on all four runtimes (rs/ruby/ts/py produce Ŭ/W/W). The three doubled-backslash rows are literal-backslash + raw text — NOT the escape — so they must not fire on Ў; the dormancy pins record exactly that (verified rs = ruby = ts = py).

Deviations from the brief, with evidence:

  • The brief expected output changes → golden regen. Rows-only change: the rs golden capture regenerates byte-identical on all 12,337 pre-existing rows (the only recapture delta is the pre-existing alalc-vai addition from corpus growth, unrelated to this PR). No golden update needed.
  • The brief said "pin one per map" assuming all six rows were true escapes; three are doubled-backslash spellings whose cross-runtime decoding DIVERGES (rs treats the row as literal \U040E, ruby effectively matches "U040E" — probed directly), so no test row can pass both. Pinned the dormancy instead.

Item 6 — spec section. docs/isc-wf-1.md §9 enumerates the four canonical-adopted rs grammar extensions (ruleset_decl, use_directive, languages_decl, the when guard) with exact semantics. docs/parity/README.md B.2 gets a G6 errata (fn-literal keyword rows resolved everywhere; the ruleset-omission note was already stale — rs joined ruleset in the G7 round).

Gates

  • iscrs test over the six maps: 172 passed, 0 failed
  • rs golden capture zero-delta (12,337/12,337)
  • ruby golden gate (tools/isc_wf1_golden_gate.rb): 12,337/12,337, exit 0
  • ts golden additivity + full suite: 394 passed, 0 failed (runtime PRs interscript-rs#21 / interscript-ts#140 / interscript-ruby#858)

…sions

Item 4: embedded test rows for the six bel/rus maps carrying capital-U escapes. gki-bel/mvd-bel/odni-bel spell the true "\U040E" escape (decodes to Ў; the generated ts/py parsers used to carry the literal "U040E" text and the row never fired) - pinned with "Ў" rows verified on all four runtimes (Ŭ/W/W). alalc-bel/bgnpcgn-bel/mvd-rus spell the row with a DOUBLED backslash - literal backslash + raw U040E text, not the escape - pinned with a "Ў" -> "Ў" dormancy row (the corrupted row must not fire on Ў; ruby and rs agree). No transliteration output changes: the rs golden capture replays 12,337/12,337 rows byte-identically.

Item 6: spec section 9 enumerates the four canonical-adopted rs grammar extensions (ruleset_decl, use_directive, languages_decl, the when guard) with exact semantics. B.2 of the parity README gets a G6 errata: the six fn-literal keyword rows are resolved (ruby's table adopted everywhere) and the ruleset row was already stale (rs joined ruleset in the G7 round).
@ronaldtse
ronaldtse merged commit 36d3068 into main Nov 14, 2026
12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant