@feigi/fleet-ctl 3.23.36 → 3.23.38
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/package.json +1 -1
- package/scripts/cell-readout.mjs +38 -11
- package/scripts/main-checkout.mjs +1 -1
- package/scripts/member-write-guard.mjs +1 -1
- package/scripts/repo-root.mjs +1 -1
- package/skills/run-team/references/reaping.md +1 -1
- package/scripts/agent-test-dir-prose.test.mjs +0 -59
- package/scripts/agent-test.test.mjs +0 -1640
- package/scripts/ambient-git-vars-mjs-prose.test.mjs +0 -397
- package/scripts/ambient-git-vars-prose.test.mjs +0 -192
- package/scripts/ancestry-check-position-prose.test.mjs +0 -175
- package/scripts/apply-ruleset.test.mjs +0 -421
- package/scripts/arg-header-probes-prose.test.mjs +0 -394
- package/scripts/arg.test.mjs +0 -1354
- package/scripts/board-cli.test.mjs +0 -710
- package/scripts/board-identity.test.mjs +0 -429
- package/scripts/board-liveness-banner.test.mjs +0 -119
- package/scripts/board-prev-shape.test.mjs +0 -228
- package/scripts/board-tryparse-null.test.mjs +0 -119
- package/scripts/board.test.mjs +0 -3121
- package/scripts/brief-after-cut-prose.test.mjs +0 -53
- package/scripts/candidates-exit3-prose.test.mjs +0 -302
- package/scripts/candidates.test.mjs +0 -2312
- package/scripts/caveat-unification-prose.test.mjs +0 -98
- package/scripts/cell-readout.test.mjs +0 -615
- package/scripts/ci-comment-rot-prose.test.mjs +0 -2025
- package/scripts/ci-completes-premise-prose.test.mjs +0 -257
- package/scripts/ci-state-prose.test.mjs +0 -113
- package/scripts/ci-state.test.mjs +0 -1933
- package/scripts/ci-vacuous-green.test.mjs +0 -368
- package/scripts/citation-sweep-prose.test.mjs +0 -1712
- package/scripts/claim-lifecycle.test.mjs +0 -187
- package/scripts/claim-ticket.test.mjs +0 -1122
- package/scripts/closing-keyword-prose.test.mjs +0 -193
- package/scripts/cockpit-launch-idempotent-prose.test.mjs +0 -165
- package/scripts/cockpit-tick-nonblocking.test.mjs +0 -210
- package/scripts/compute-board.test.mjs +0 -1214
- package/scripts/compute-spend.test.mjs +0 -517
- package/scripts/consumer-technology-prose.test.mjs +0 -126
- package/scripts/cross-repo-citation-prose.test.mjs +0 -155
- package/scripts/cwd-isolation-pins.mjs +0 -83
- package/scripts/dedupe-guard-prose.test.mjs +0 -243
- package/scripts/delete-merged-branch.test.mjs +0 -576
- package/scripts/derive-testcmd.test.mjs +0 -595
- package/scripts/diff-stats.test.mjs +0 -404
- package/scripts/dispatch-block-golden-prose.test.mjs +0 -1130
- package/scripts/dispatch-block-pins-prose.test.mjs +0 -615
- package/scripts/dispositions-check.test.mjs +0 -936
- package/scripts/dispositions-gate-prose.test.mjs +0 -129
- package/scripts/drop-merged-label.test.mjs +0 -479
- package/scripts/exec-stub.mjs +0 -93
- package/scripts/exec-stub.test.mjs +0 -69
- package/scripts/expected-jobs-prose.test.mjs +0 -80
- package/scripts/fan-out-authorization-prose.test.mjs +0 -80
- package/scripts/filing-bar-seat-prose.test.mjs +0 -242
- package/scripts/filing-claim-bar-prose.test.mjs +0 -166
- package/scripts/finisher-died-prose.test.mjs +0 -31
- package/scripts/finisher-dispatch-premise-prose.test.mjs +0 -312
- package/scripts/finisher-halt-prose.test.mjs +0 -106
- package/scripts/finisher-name-prose.test.mjs +0 -134
- package/scripts/finisher-own-tree-prose.test.mjs +0 -299
- package/scripts/finisher-pin-race-prose.test.mjs +0 -339
- package/scripts/finisher-test-verdict-prose.test.mjs +0 -102
- package/scripts/fix-applier-correction-rules-prose.test.mjs +0 -202
- package/scripts/fix-applier-result-file-prose.test.mjs +0 -149
- package/scripts/fleet-bootstrap.test.mjs +0 -75
- package/scripts/fleet-heartbeat.test.mjs +0 -450
- package/scripts/fleet-provenance.test.mjs +0 -138
- package/scripts/fleet-run.test.mjs +0 -90
- package/scripts/fleet-state.test.mjs +0 -460
- package/scripts/fleet-tick-prose.test.mjs +0 -300
- package/scripts/fleet-tick.test.mjs +0 -2444
- package/scripts/force-with-lease-reason-prose.test.mjs +0 -184
- package/scripts/frontmatter-check.test.mjs +0 -436
- package/scripts/git-env.test.mjs +0 -149
- package/scripts/git-status-untracked-mode-sweep.test.mjs +0 -216
- package/scripts/gone-walk.test.mjs +0 -213
- package/scripts/gp-sep-invariant-prose.test.mjs +0 -188
- package/scripts/head-moved-after-label-prose.test.mjs +0 -233
- package/scripts/immutable-body-claim-prose.test.mjs +0 -204
- package/scripts/implementer-model-tier.test.mjs +0 -235
- package/scripts/inflight-citation-prose.test.mjs +0 -144
- package/scripts/inflight-exit2-prose.test.mjs +0 -81
- package/scripts/inflight-probe-guard-prose.test.mjs +0 -109
- package/scripts/inflight-scan-scope-prose.test.mjs +0 -57
- package/scripts/inflight.test.mjs +0 -2882
- package/scripts/injection-control-prose.test.mjs +0 -404
- package/scripts/install-root-audit.test.mjs +0 -464
- package/scripts/instrument-check-prose.test.mjs +0 -235
- package/scripts/instruments.test.mjs +0 -787
- package/scripts/is-cli.test.mjs +0 -99
- package/scripts/issue-tracker-prose.test.mjs +0 -71
- package/scripts/json.test.mjs +0 -364
- package/scripts/ledger-dispatch.test.mjs +0 -927
- package/scripts/ledger-grammar.test.mjs +0 -227
- package/scripts/ledger-read-require-file.test.mjs +0 -197
- package/scripts/ledger-rotate.test.mjs +0 -142
- package/scripts/ledger.test.mjs +0 -3436
- package/scripts/line-distance-prose.test.mjs +0 -245
- package/scripts/liveness-rationale-prose.test.mjs +0 -83
- package/scripts/locale-pin-prose.test.mjs +0 -189
- package/scripts/main-checkout.test.mjs +0 -332
- package/scripts/member-outcome-prose.test.mjs +0 -167
- package/scripts/member-outcomes-header.test.mjs +0 -129
- package/scripts/member-outcomes-prose.test.mjs +0 -63
- package/scripts/member-outcomes.test.mjs +0 -638
- package/scripts/member-prompt-prose.test.mjs +0 -165
- package/scripts/member-record.test.mjs +0 -541
- package/scripts/member-write-guard.test.mjs +0 -214
- package/scripts/merge-commit-probe-prose.test.mjs +0 -122
- package/scripts/merge-gate.test.mjs +0 -613
- package/scripts/muted-git-guard-sweep.test.mjs +0 -493
- package/scripts/net-ssh-precedence-prose.test.mjs +0 -91
- package/scripts/net-watchdog-prose.test.mjs +0 -79
- package/scripts/net.test.mjs +0 -211
- package/scripts/no-ci-gate-prose.test.mjs +0 -108
- package/scripts/no-undo-audit.test.mjs +0 -3961
- package/scripts/no-wave.test.mjs +0 -138
- package/scripts/node-floor-sweep.test.mjs +0 -611
- package/scripts/pointer-target-prose.test.mjs +0 -186
- package/scripts/pr-cost.test.mjs +0 -521
- package/scripts/pr-overlap.test.mjs +0 -692
- package/scripts/printf-die-sweep.test.mjs +0 -384
- package/scripts/prompt-renderer.mjs +0 -40
- package/scripts/prompt-renderer.test.mjs +0 -68
- package/scripts/prose-pin.mjs +0 -807
- package/scripts/prose-pin.test.mjs +0 -1034
- package/scripts/prove-merge.test.mjs +0 -995
- package/scripts/pull-table-relabel-prose.test.mjs +0 -60
- package/scripts/quiet-payload-prose.test.mjs +0 -222
- package/scripts/reap-cherry-check-prose.test.mjs +0 -80
- package/scripts/reap.test.mjs +0 -4432
- package/scripts/reaping-prose.test.mjs +0 -239
- package/scripts/reaping-reap-authorizers-prose.test.mjs +0 -112
- package/scripts/reaping-release-authorizers-prose.test.mjs +0 -85
- package/scripts/recipe-derivation-prose.test.mjs +0 -112
- package/scripts/recipe-prove.test.mjs +0 -1042
- package/scripts/refuter-scratch-prose.test.mjs +0 -135
- package/scripts/release-label-required-prose.test.mjs +0 -71
- package/scripts/release-label-split-prose.test.mjs +0 -184
- package/scripts/release-ticket-reap-pairing-prose.test.mjs +0 -63
- package/scripts/release-ticket.test.mjs +0 -5099
- package/scripts/renovate-release-contract.test.mjs +0 -379
- package/scripts/repo-root.test.mjs +0 -873
- package/scripts/reporter-output-rule-prose.test.mjs +0 -208
- package/scripts/review-agent-report-only.test.mjs +0 -49
- package/scripts/review-core-booking.test.mjs +0 -35
- package/scripts/review-core-cwd-isolation.test.mjs +0 -374
- package/scripts/review-core-reads.test.mjs +0 -478
- package/scripts/review-core-refuter-scratch.test.mjs +0 -148
- package/scripts/review-core-scratch-refusal.test.mjs +0 -64
- package/scripts/review-core-shared-test-run.test.mjs +0 -275
- package/scripts/review-core-snapshot-path.test.mjs +0 -830
- package/scripts/review-core-specialist-scratch.test.mjs +0 -191
- package/scripts/review-core-testcmd.test.mjs +0 -335
- package/scripts/review-core-unrun.test.mjs +0 -566
- package/scripts/review-core-unverified-discriminant.test.mjs +0 -208
- package/scripts/review-eval-resolver-path.test.mjs +0 -55
- package/scripts/review-host-fixture.mjs +0 -89
- package/scripts/review-in-run-retry.test.mjs +0 -211
- package/scripts/review-no-vendor-dispatch.test.mjs +0 -45
- package/scripts/review-path-default.test.mjs +0 -796
- package/scripts/review-runner-spawns.test.mjs +0 -95
- package/scripts/review-runner.test.mjs +0 -166
- package/scripts/review-tier-audit.test.mjs +0 -140
- package/scripts/run-artifact-pr-prose.test.mjs +0 -341
- package/scripts/run-merge-bot-prose.test.mjs +0 -1413
- package/scripts/run-team-local-hook.test.mjs +0 -442
- package/scripts/scratch-partition-invariant-prose.test.mjs +0 -74
- package/scripts/scripts-path-citation-sweep.test.mjs +0 -73
- package/scripts/section-between-guard-prose.test.mjs +0 -379
- package/scripts/select-dimensions.test.mjs +0 -674
- package/scripts/shared-refusal.test.mjs +0 -388
- package/scripts/shell-traps-prose.test.mjs +0 -504
- package/scripts/shortlist.test.mjs +0 -380
- package/scripts/slow-transport.mjs +0 -72
- package/scripts/snapshot-repo.test.mjs +0 -1006
- package/scripts/snapshot-runner-audience-prose.test.mjs +0 -130
- package/scripts/spend-view.test.mjs +0 -312
- package/scripts/staleness-qualifier-prose.test.mjs +0 -251
- package/scripts/staleness-verdict-prose.test.mjs +0 -142
- package/scripts/staleness.test.mjs +0 -893
- package/scripts/stalled-member-liveness-prose.test.mjs +0 -219
- package/scripts/stash-prohibition-prose.test.mjs +0 -313
- package/scripts/strip-comments.mjs +0 -44
- package/scripts/temp-dir.mjs +0 -47
- package/scripts/ticket-router.test.mjs +0 -981
- package/scripts/tier-check.test.mjs +0 -828
- package/scripts/tier-outcomes-header.test.mjs +0 -518
- package/scripts/tier-outcomes.test.mjs +0 -572
- package/scripts/tier-roles.test.mjs +0 -535
- package/scripts/tracker-block-copy-prose.test.mjs +0 -288
- package/scripts/triage-labels-gloss-prose.test.mjs +0 -97
- package/scripts/unattended-git-sweep.test.mjs +0 -142
- package/scripts/verify-sha-exit2-prose.test.mjs +0 -269
- package/scripts/verify-sha.test.mjs +0 -692
- package/scripts/watcher-degraded-block.test.mjs +0 -369
- package/scripts/watcher-key-prose.test.mjs +0 -106
- package/scripts/within-run-pair-prose.test.mjs +0 -212
- package/scripts/worktree-audit.test.mjs +0 -1408
- package/scripts/worktree-head-prose.test.mjs +0 -276
- package/scripts/worktree-listing-sweep.test.mjs +0 -217
- package/scripts/worktree-model-prose.test.mjs +0 -216
- package/scripts/worktree-porcelain.mjs +0 -35
- package/scripts/worktree-porcelain.test.mjs +0 -102
package/scripts/prose-pin.mjs
DELETED
|
@@ -1,807 +0,0 @@
|
|
|
1
|
-
// THE PIN CONVENTION (#491), for every new prose pin in this directory.
|
|
2
|
-
// Existing pins move onto it only where a hole in them is measured.
|
|
3
|
-
//
|
|
4
|
-
// A pin holding a claim asserts ONE contiguous `phrase()` span: the smallest
|
|
5
|
-
// clause that makes the claim, never independent keywords. Keywords each
|
|
6
|
-
// satisfied somewhere in the slice stay green when the sentence carrying them
|
|
7
|
-
// is gutted and its words restated beside it with the meaning inverted: the
|
|
8
|
-
// slice bounds WHERE the words are, only a span binds them to each other.
|
|
9
|
-
// Measured on ci-state-prose.test.mjs, where three independent checks (two
|
|
10
|
-
// regexes and a substring match) stayed green
|
|
11
|
-
// under exactly that restatement inside the pinned paragraph. A literal the
|
|
12
|
-
// source owns (a reason string read out of the script) may be interpolated
|
|
13
|
-
// into the span, so renaming it there reddens the pin too.
|
|
14
|
-
//
|
|
15
|
-
// A reflow stays green, since `phrase()` joins words on `\s+`; so does a `**`
|
|
16
|
-
// move, where the slice is matched through `unemphasized()`. A reword inside
|
|
17
|
-
// the span goes red, and that is accepted as the loud direction: the assertion
|
|
18
|
-
// message names what the span protects, so whoever rewords it re-pins
|
|
19
|
-
// deliberately. `sentences()` may locate a sentence; it does not replace the
|
|
20
|
-
// span.
|
|
21
|
-
//
|
|
22
|
-
// THE CEILING: a negating sentence added beside an intact span stays green.
|
|
23
|
-
// No tighter regex closes that; it is the limit of pinning prose by matching it.
|
|
24
|
-
//
|
|
25
|
-
// A list item is sliced with `bullet()`, never with a bare `between()` on a
|
|
26
|
-
// fixed end anchor, for the same reason one level up: every sibling item
|
|
27
|
-
// inserted between the anchors joins that slice and can carry the span the
|
|
28
|
-
// real item lost.
|
|
29
|
-
|
|
30
|
-
import assert from "node:assert/strict";
|
|
31
|
-
|
|
32
|
-
// How many times `needle` occurs in `haystack`, overlaps included — two
|
|
33
|
-
// overlapping hits are two places a slice could start, which is exactly the
|
|
34
|
-
// ambiguity the caller below refuses. A count, not the offsets: the caller
|
|
35
|
-
// needs only "more than one" and the number to say so with. Cursor-bounded
|
|
36
|
-
// rather than chained off the previous hit, so an anchor that strips to ""
|
|
37
|
-
// terminates instead of spinning on `indexOf("", i)`'s clamped return.
|
|
38
|
-
function occurrenceCount(haystack, needle) {
|
|
39
|
-
let n = 0;
|
|
40
|
-
for (let i = 0; i <= haystack.length; ) {
|
|
41
|
-
const hit = haystack.indexOf(needle, i);
|
|
42
|
-
if (hit === -1) break;
|
|
43
|
-
n += 1;
|
|
44
|
-
i = hit + 1;
|
|
45
|
-
}
|
|
46
|
-
return n;
|
|
47
|
-
}
|
|
48
|
-
|
|
49
|
-
// Bound at BOTH ends: an unbounded end lets a later, unrelated occurrence of the
|
|
50
|
-
// same phrase satisfy the assertion with the real clause deleted.
|
|
51
|
-
//
|
|
52
|
-
// `emphasisTolerant` mirrors the option `anchorAt` and `quoteBlock` already
|
|
53
|
-
// carry — same name, same default of off — and reuses their `unemphasized`/
|
|
54
|
-
// `rawOffset` pair rather than carrying a second stripper. On, BOTH the anchors
|
|
55
|
-
// and the document are read through `unemphasized`, so `**` may be removed,
|
|
56
|
-
// added, narrowed, widened, or moved to another word boundary anywhere inside
|
|
57
|
-
// an anchor without moving the slice. Stripping the ANCHOR too is what
|
|
58
|
-
// `anchorAt`'s own tolerant mode does not do, and is the direction this family
|
|
59
|
-
// needs: its anchors are typed WITH `**` when the source is, so a document that
|
|
60
|
-
// loses the emphasis is the common break.
|
|
61
|
-
//
|
|
62
|
-
// BOTH anchors must then resolve EXACTLY ONCE when tolerant — the standard
|
|
63
|
-
// `anchorAt` and `quoteBlock` already hold for their one anchor, and the price
|
|
64
|
-
// of tolerance here, paid at both ends alike (`occurrenceCount` below is
|
|
65
|
-
// shared by both guards). Stripping `**` can only ADD matches to a search over
|
|
66
|
-
// the stripped view, never remove one the literal anchor had, so a tolerant
|
|
67
|
-
// anchor can resolve somewhere the literal one never could: on a text-identical
|
|
68
|
-
// phrase sitting in a different emphasis state elsewhere in the document, or —
|
|
69
|
-
// when an anchor literal itself carries a `*` run that does not collapse to a
|
|
70
|
-
// clean `**`-pair count — on a shorter, less specific phrase the literal anchor
|
|
71
|
-
// never matched at all, which can make a tolerant match genuinely WIDER than
|
|
72
|
-
// the literal one, not merely earlier. Either direction is invisible to every
|
|
73
|
-
// assertion inside the slice: at the START, the pinned words are all present,
|
|
74
|
-
// sourced from the wrong copy; at the END, a slice that lands short silently
|
|
75
|
-
// DELETES content, which can flip a `doesNotMatch` pin from correctly RED to
|
|
76
|
-
// falsely GREEN — the same false-green class a wrong START does, just reached
|
|
77
|
-
// from the other side. Both throw instead, for the reason a missing anchor
|
|
78
|
-
// throws rather than widening to the whole file.
|
|
79
|
-
//
|
|
80
|
-
// Only the TOLERANT path pays that price. The literal path's END anchor keeps
|
|
81
|
-
// first-hit-after-the-start and stays non-unique ON PURPOSE — an end bound's
|
|
82
|
-
// job there is to be the next one, not the only one — and two of this
|
|
83
|
-
// directory's real end anchors depend on exactly that:
|
|
84
|
-
// `snapshot-runner-audience-prose.test.mjs`'s `"\n- **"` means "wherever the
|
|
85
|
-
// next bullet starts", and `instrument-check-prose.test.mjs`'s `"**Exit 0 is
|
|
86
|
-
// the only code"` names a sentence `run-team/SKILL.md` states twice, once bold
|
|
87
|
-
// and once inside a blockquote. Neither callsite turns `emphasisTolerant` on,
|
|
88
|
-
// so the guard above never sees them; an exactly-once literal-mode end bound
|
|
89
|
-
// would refuse both while preventing no widening that exists today.
|
|
90
|
-
//
|
|
91
|
-
// Matching stays a literal `indexOf` over the stripped view, NOT `phrase()` as
|
|
92
|
-
// `anchorAt` uses: #1492's own ticket measured that trade. `phrase()` is
|
|
93
|
-
// `s.trim().split(/\s+/)`, which discards a leading newline — and `between`
|
|
94
|
-
// anchors carry load-bearing ones, e.g. `fix-applier-correction-rules-prose
|
|
95
|
-
// .test.mjs`'s `PROMPT_END`, `"\n**Put the standing CI facts"`: the leading
|
|
96
|
-
// `\n` is what ties that anchor to the phrase's own line start rather than to
|
|
97
|
-
// any mid-line occurrence of the same words. `phrase()`'s `.trim()` would
|
|
98
|
-
// strip that newline before matching, silently loosening the anchor. A
|
|
99
|
-
// tolerant mode built on `phrase()` would trade this family's false reds for
|
|
100
|
-
// that false green. Emphasis tolerance and whitespace tolerance are separate
|
|
101
|
-
// axes; this option moves only the first.
|
|
102
|
-
//
|
|
103
|
-
// The returned slice is raw bytes with `**` intact — `rawOffset` maps each
|
|
104
|
-
// endpoint back and stops BEFORE the marker, so a tolerant slice keeps the
|
|
105
|
-
// emphasis its caller goes on to compare.
|
|
106
|
-
export function between(text, from, to, what, { emphasisTolerant = false } = {}) {
|
|
107
|
-
const haystack = emphasisTolerant ? unemphasized(text) : text;
|
|
108
|
-
const start = emphasisTolerant ? unemphasized(from) : from;
|
|
109
|
-
const stop = emphasisTolerant ? unemphasized(to) : to;
|
|
110
|
-
const at = haystack.indexOf(start);
|
|
111
|
-
assert.notEqual(at, -1, `${what} no longer contains "${from}" — update this test`);
|
|
112
|
-
if (emphasisTolerant) {
|
|
113
|
-
const hits = occurrenceCount(haystack, start);
|
|
114
|
-
assert.equal(
|
|
115
|
-
hits,
|
|
116
|
-
1,
|
|
117
|
-
`${what}: emphasis-tolerant slice anchor "${from}" occurs ${hits} times once \`**\` is ignored — the slice would start at the first and pull in what the literal anchor excluded; narrow the anchor or drop emphasisTolerant`,
|
|
118
|
-
);
|
|
119
|
-
}
|
|
120
|
-
const searchFrom = at + start.length;
|
|
121
|
-
const end = haystack.indexOf(stop, searchFrom);
|
|
122
|
-
assert.notEqual(end, -1, `${what} no longer contains "${to}" after "${from}" — update this test`);
|
|
123
|
-
if (emphasisTolerant) {
|
|
124
|
-
const hits = occurrenceCount(haystack.slice(searchFrom), stop);
|
|
125
|
-
assert.equal(
|
|
126
|
-
hits,
|
|
127
|
-
1,
|
|
128
|
-
`${what}: emphasis-tolerant slice end anchor "${to}" occurs ${hits} times after the start once \`**\` is ignored — the slice would end at the first and could drop or admit content the literal anchor did not; narrow the anchor or drop emphasisTolerant`,
|
|
129
|
-
);
|
|
130
|
-
}
|
|
131
|
-
return emphasisTolerant ? text.slice(rawOffset(text, at), rawOffset(text, end)) : text.slice(at, end);
|
|
132
|
-
}
|
|
133
|
-
|
|
134
|
-
// Regex metacharacters escaped so a caller's literal stays a literal. One
|
|
135
|
-
// definition because both consumers below take arbitrary caller text — a
|
|
136
|
-
// second copy of this class is how the two drift apart.
|
|
137
|
-
const escapeRe = (s) => s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
|
|
138
|
-
|
|
139
|
-
// `\s+` between every word, never a literal space — prose this matches against
|
|
140
|
-
// is often hard-wrapped, so any inter-word space in the source may be a
|
|
141
|
-
// newline plus indent. Regex metacharacters in the phrase are escaped first.
|
|
142
|
-
export const phrase = (s) => new RegExp(s.trim().split(/\s+/).map(escapeRe).join("\\s+"));
|
|
143
|
-
|
|
144
|
-
// The paragraph carrying a rule, and no more of the document than that. A
|
|
145
|
-
// positive regex matched over a whole file is satisfiable from OUTSIDE the
|
|
146
|
-
// clause it guards — a stray copy of the wording left anywhere else buys the
|
|
147
|
-
// pass — so a prose pin's slice bound is the pin, and the regex is the easy
|
|
148
|
-
// half. Ends at the blank line, because a paragraph boundary is the one
|
|
149
|
-
// bound a rewrap cannot move; the ANCHOR goes through `phrase()` for the
|
|
150
|
-
// mirror-image reason, since `between`'s literal `indexOf` would break on a
|
|
151
|
-
// rewrap the pinned clause itself survives, turning a reflow into a red.
|
|
152
|
-
// A missing anchor throws rather than widening: silently falling back to the
|
|
153
|
-
// whole document is the false green this bound exists to prevent.
|
|
154
|
-
//
|
|
155
|
-
// The blank line is matched as `\n[ \t]*\n`, never the literal `\n\n`: an
|
|
156
|
-
// editor that keeps a list item's indent on the line between two paragraphs
|
|
157
|
-
// writes `\n \n`, which a literal search does not see — the slice then
|
|
158
|
-
// widens silently into the next paragraph, and a decoy there buys the green
|
|
159
|
-
// this bound exists to deny. No `*.md` in this repo carries a whitespace-only
|
|
160
|
-
// line today and nothing enforces that — no `.editorconfig`, no prettier or
|
|
161
|
-
// markdownlint config, no CI check — and every paragraph pinned so far is
|
|
162
|
-
// indented list content, which is exactly where an editor produces that shape.
|
|
163
|
-
//
|
|
164
|
-
// The anchor must match EXACTLY ONCE — see `anchorAt`, which holds that half.
|
|
165
|
-
//
|
|
166
|
-
// The single definition of a paragraph bound in this directory. It was
|
|
167
|
-
// extracted (#823) while pins elsewhere still hand-rolled this same
|
|
168
|
-
// search-to-blank-line slicer locally, every copy carrying both false greens
|
|
169
|
-
// above by construction; #1372 migrated them onto it. A pin needing this bound
|
|
170
|
-
// imports it — a local copy is the defect, not a style choice.
|
|
171
|
-
export function paragraph(text, anchor, what, options) {
|
|
172
|
-
const rest = text.slice(anchorAt(text, anchor, what, options));
|
|
173
|
-
const end = rest.search(/\n[ \t]*\n/);
|
|
174
|
-
return end === -1 ? rest : rest.slice(0, end);
|
|
175
|
-
}
|
|
176
|
-
|
|
177
|
-
// One list item: from `from` to where markdown ends the item, and never past
|
|
178
|
-
// `to`. `between` alone on a fixed end anchor takes every item between the two
|
|
179
|
-
// anchors, so a sibling inserted there joins the slice and carries a pin the
|
|
180
|
-
// real item lost. Measured on run-team's `no-ci` edge (#747, #491): the ruling
|
|
181
|
-
// clause gutted, then restated in an inserted sibling bullet, stayed green. `to`
|
|
182
|
-
// stays as the outer bound and as the "moved — update this test" tripwire
|
|
183
|
-
// `between` already raises.
|
|
184
|
-
//
|
|
185
|
-
// A line belongs to the item only if it is indented to the item's CONTENT
|
|
186
|
-
// column — past the marker and the gap after it, CommonMark's rule — or is a
|
|
187
|
-
// lazy continuation: a plain line straight after the item's text, which
|
|
188
|
-
// markdown renders inside the item's paragraph. A shallower line ends the item
|
|
189
|
-
// when it opens a list marker (a sibling, whatever its marker: `-`, `*`, `+`,
|
|
190
|
-
// `1.`, `1)`), when it opens a block that interrupts a paragraph (an ATX
|
|
191
|
-
// heading, `>`, a fence, a thematic break — not an HTML block, which CommonMark
|
|
192
|
-
// also lets interrupt one: a shallower `<div>`, `</div>`, `<!--`, `<script>`,
|
|
193
|
-
// `<?` or `<!DOCTYPE` line ends the item in CommonMark, yet this reads it as
|
|
194
|
-
// lazy and keeps it; a `<span>`-style line interrupts in neither), or when a
|
|
195
|
-
// blank line precedes it.
|
|
196
|
-
// Only the first ended it before #2077, so a restatement in a paragraph after
|
|
197
|
-
// the item, or under a new heading, stayed inside the slice and held the pin.
|
|
198
|
-
// The marker and interrupt cases count only short of the item's MARKER column
|
|
199
|
-
// + 4: from there on, with no blank line before it, the line is lazy
|
|
200
|
-
// continuation, since 4+ columns past its container a line cannot open a block
|
|
201
|
-
// or a list item and indented code cannot interrupt a paragraph. That zone
|
|
202
|
-
// exists only when the content column is past marker column + 4 — a `100.`
|
|
203
|
-
// marker, or a 4-space gap (#2100); a blank line still ends the item there.
|
|
204
|
-
// Not modelled: which block the previous line was, so a shallower plain line
|
|
205
|
-
// straight after the item's own heading, fence, thematic break, indented code
|
|
206
|
-
// or HTML block — none of them a paragraph a lazy line could continue — reads
|
|
207
|
-
// as lazy and stays in; so does a shallower line inside the item's own
|
|
208
|
-
// still-open fence, where CommonMark ends the fence and the item both; and so
|
|
209
|
-
// does any shallower line at marker column + 4 or deeper in that zone.
|
|
210
|
-
// Nor the parent's content column: marker column + 4 stands in for it + 4,
|
|
211
|
-
// exact while the marker sits AT that column, as in any normal list. A marker
|
|
212
|
-
// indented 1-3 columns past it (` 100. **Item**` at top level) still ends
|
|
213
|
-
// the item at a heading or marker line CommonMark reads as lazy.
|
|
214
|
-
//
|
|
215
|
-
// Columns are visual, a tab advancing to the next multiple of 4 (CommonMark's
|
|
216
|
-
// tab stop): counted in characters, a one-tab item read a two-space sibling as
|
|
217
|
-
// a deeper child (#2077). The item's line is the one `from`'s first
|
|
218
|
-
// non-whitespace character sits on, and the walk starts after `from`'s trailing
|
|
219
|
-
// whitespace, so a leading-`\n` anchor — load-bearing in `between` — measures
|
|
220
|
-
// its own line and not the one before it, and a trailing-`\n` anchor does not
|
|
221
|
-
// eat the line break the next line's check needs (#2077). `from` must occur
|
|
222
|
-
// exactly once: a list-item anchor is a start anchor, and a second copy would
|
|
223
|
-
// silently decide which item is read.
|
|
224
|
-
//
|
|
225
|
-
// Returns raw bytes starting with `from` in full, like `between`. When a line
|
|
226
|
-
// of the text ends the item, the blank lines before that line are dropped;
|
|
227
|
-
// when `to` ends it, the slice comes back as `between` gives it, trailing blank
|
|
228
|
-
// lines and all (run-team's no-ci edge ends in `\n\n`). A caller flattens or
|
|
229
|
-
// matches through `phrase()`.
|
|
230
|
-
//
|
|
231
|
-
// The single definition of a list-item bound in this directory (#491), moved
|
|
232
|
-
// out of no-ci-gate-prose.test.mjs, where #747 had clamped a local copy.
|
|
233
|
-
export function bullet(text, from, to, what) {
|
|
234
|
-
const slice = between(text, from, to, what);
|
|
235
|
-
const hits = occurrenceCount(text, from);
|
|
236
|
-
assert.equal(hits, 1, `${what}: list-item anchor "${from}" occurs ${hits} times — the item would be read off the first copy; narrow the anchor`);
|
|
237
|
-
const at = text.indexOf(from) + (from.length - from.trimStart().length);
|
|
238
|
-
const lineStart = text.lastIndexOf("\n", at - 1) + 1;
|
|
239
|
-
const lineEnd = text.indexOf("\n", at);
|
|
240
|
-
const itemLine = text.slice(lineStart, lineEnd === -1 ? text.length : lineEnd);
|
|
241
|
-
const content = contentColumn(itemLine);
|
|
242
|
-
const lazyFrom = columnOf(itemLine.match(/^[ \t]*/)[0]) + 4;
|
|
243
|
-
let kept = slice.indexOf("\n", from.trimEnd().length);
|
|
244
|
-
let blankBefore = false;
|
|
245
|
-
for (let nl = kept; nl !== -1; ) {
|
|
246
|
-
const next = slice.indexOf("\n", nl + 1);
|
|
247
|
-
const line = slice.slice(nl + 1, next === -1 ? slice.length : next);
|
|
248
|
-
const indent = line.match(/^[ \t]*/)[0];
|
|
249
|
-
const col = columnOf(indent);
|
|
250
|
-
if (indent.length === line.length) {
|
|
251
|
-
blankBefore = true;
|
|
252
|
-
} else if (col < content && (blankBefore || (col < lazyFrom && (ITEM_MARKER.test(line) || PARAGRAPH_INTERRUPT.test(line))))) {
|
|
253
|
-
return slice.slice(0, Math.max(kept, from.length));
|
|
254
|
-
} else {
|
|
255
|
-
blankBefore = false;
|
|
256
|
-
kept = next;
|
|
257
|
-
}
|
|
258
|
-
nl = next;
|
|
259
|
-
}
|
|
260
|
-
return slice;
|
|
261
|
-
}
|
|
262
|
-
|
|
263
|
-
// Visual width of leading whitespace, a tab advancing to the next multiple of 4.
|
|
264
|
-
function columnOf(whitespace) {
|
|
265
|
-
let col = 0;
|
|
266
|
-
for (const c of whitespace) col = c === "\t" ? col + 4 - (col % 4) : col + 1;
|
|
267
|
-
return col;
|
|
268
|
-
}
|
|
269
|
-
|
|
270
|
-
// Where a list item's text starts: past the marker and its gap, or one column
|
|
271
|
-
// past the marker when the gap is over 4 (the rest is an indented code block).
|
|
272
|
-
// A line with no marker has no content column of its own, so anything deeper
|
|
273
|
-
// than its indent counts as its child.
|
|
274
|
-
function contentColumn(line) {
|
|
275
|
-
const item = line.match(/^([ \t]*(?:[-*+]|\d+[.)]))([ \t]+)/);
|
|
276
|
-
if (!item) return columnOf(line.match(/^[ \t]*/)[0]) + 1;
|
|
277
|
-
const marker = columnOf(item[1]);
|
|
278
|
-
const gap = columnOf(item[1] + item[2]) - marker;
|
|
279
|
-
return gap > 4 ? marker + 1 : marker + gap;
|
|
280
|
-
}
|
|
281
|
-
|
|
282
|
-
// A marker with nothing after it on its own line (content on the next,
|
|
283
|
-
// indented line) is still a valid, separate list item — `(?:[ \t]|$)`, not
|
|
284
|
-
// only `[ \t]`, so a bare `-`/`2.` sibling is recognised too. Reviewed PR
|
|
285
|
-
// #2091, hand-dispatched review: `bullet('- **A** body\n-\n restated
|
|
286
|
-
// claim\nEND','- **A**','END')` kept the bare-marker sibling inside the
|
|
287
|
-
// slice under the narrower form — the same #747/#2077 escape this bound
|
|
288
|
-
// exists to close, just via a marker shape the sibling regex didn't cover.
|
|
289
|
-
const ITEM_MARKER = /^[ \t]*(?:[-*+]|\d+[.)])(?:[ \t]|$)/;
|
|
290
|
-
// A thematic break may space its repeated character out (`- - -`, `_ _ _`,
|
|
291
|
-
// `* * *`), not only run it together — CommonMark's own rule. The `_`/`*`
|
|
292
|
-
// forms have no marker regex to fall back on (unlike `-`, which ITEM_MARKER
|
|
293
|
-
// already catches), so an unspaced-only rule here missed them. Same review.
|
|
294
|
-
const PARAGRAPH_INTERRUPT = /^[ \t]*(?:#{1,6}(?:[ \t]|$)|>|```|~~~|(?:-[ \t]*){3,}$|(?:\*[ \t]*){3,}$|(?:_[ \t]*){3,}$)/;
|
|
295
|
-
|
|
296
|
-
// Prose cut into sentences, for a pin that holds a claim to ONE sentence. A
|
|
297
|
-
// sentence ends at `.`, `!` or `?` followed by whitespace, with any closing
|
|
298
|
-
// marks between the two — `.**`, `.)`, `."` — so a bold lead-in, a
|
|
299
|
-
// parenthetical sentence or a quoted one ends where its reader sees it end.
|
|
300
|
-
// Without them a pinned sentence runs on into whatever follows a bold one, a
|
|
301
|
-
// shape run-team's SKILL.md opens paragraphs with throughout (#1940). A
|
|
302
|
-
// backtick is not one: a code span ending in punctuation (`git restore .`,
|
|
303
|
-
// `runSubprocess(...)`) sits mid-sentence far more often than it ends one.
|
|
304
|
-
//
|
|
305
|
-
// A period closing an abbreviation that never ends a sentence — "e.g.",
|
|
306
|
-
// "i.e.", "cf.", "viz.", "vs." — is no break, closing marks or not ("e.g.)"):
|
|
307
|
-
// splitting there scattered one sentence's words across two fragments, neither
|
|
308
|
-
// of which satisfied the pin, so a paraphrase using one was refused (#1852).
|
|
309
|
-
// "etc." is deliberately absent: it ends sentences as often as not, and reading
|
|
310
|
-
// past it would join two real sentences. A period inside a word ("ci.yml",
|
|
311
|
-
// "v1.2") is no break either, since no whitespace follows it.
|
|
312
|
-
//
|
|
313
|
-
// Only a letter or digit right before the abbreviation disqualifies it, so an
|
|
314
|
-
// italic `_e.g._` is one (#2053). A `\b` there read the `_` as a word
|
|
315
|
-
// character, so the period split the sentence mid-abbreviation — the same gap
|
|
316
|
-
// #2021/#2043 closed in citation-sweep's SENTENCE_END and jsonCountFault's
|
|
317
|
-
// re-split, whose guard this now matches. Running first does not make the
|
|
318
|
-
// narrower guard moot for a caller that re-splits a slice: a re-split can only
|
|
319
|
-
// cut a piece further, never rejoin two this function wrongly parted. The
|
|
320
|
-
// cost: a bare snake_case word ending in any of the five — `_vs.`, `_cf.`,
|
|
321
|
-
// `_viz.`, `_e.g.`, `_i.e.` — at a sentence's real end now joins the next
|
|
322
|
-
// one, the same underscore-adjacency this fix widens the guard to accept.
|
|
323
|
-
// Measured over every tracked .md/.mjs/.yml/.sh/
|
|
324
|
-
// .json file, the widening moved no split but three in fixtures of those two
|
|
325
|
-
// sibling fixes, each an italic abbreviation.
|
|
326
|
-
//
|
|
327
|
-
// The abbreviations match lowercase only, as this repo's prose spells every
|
|
328
|
-
// one of them (#1899). Folded case, a sentence genuinely ending in a word
|
|
329
|
-
// that merely shares their letters — a proper noun "Vs.", a stray "E.g." —
|
|
330
|
-
// read as the abbreviation and joined the sentence after it, the same silent
|
|
331
|
-
// merge "etc." is kept off the list to avoid. The cost runs the other way: a
|
|
332
|
-
// capitalized one anywhere — a sentence OPENING "E.g. …", a parenthetical
|
|
333
|
-
// "(Cf. …)" mid-sentence — now ends the sentence right there, not only at a
|
|
334
|
-
// sentence's own start. No file's PROSE spells any of the five capitalized,
|
|
335
|
-
// outside this ticket's own fixtures illustrating the defect (#1899 grepped
|
|
336
|
-
// the corpus).
|
|
337
|
-
//
|
|
338
|
-
// This replaces the first-period `[^.]*` scan (#1898, #1940), which is wrong
|
|
339
|
-
// both ways at once: it ends the window at every period above, and it runs
|
|
340
|
-
// straight past a `?` or `!` into the next sentence. Which of the two is the
|
|
341
|
-
// SILENT direction depends on the pin. For one that requires something to be
|
|
342
|
-
// IN the sentence, a window cut short reds on a valid rewording and a window
|
|
343
|
-
// run long lets a neighbour supply what the sentence lacks; for one that checks
|
|
344
|
-
// everything the sentence holds — citationFault's quoted spans — the two swap.
|
|
345
|
-
// So a boundary this function misjudges is silent for some caller whichever
|
|
346
|
-
// way it errs, and the calls above are the ones prose makes far more often
|
|
347
|
-
// than not. Two it gets wrong: a bolded or quoted question used mid-sentence
|
|
348
|
-
// as a name (`judge **decided?** — …`) splits, as a bare `?` always has; and a
|
|
349
|
-
// whole sentence in parentheses mid-sentence ends the outer one at its `.)`.
|
|
350
|
-
//
|
|
351
|
-
// One definition, extracted (#1940) from ci-comment-rot-prose.test.mjs, where
|
|
352
|
-
// it bounded ci.yml's comment prose alone — the same reason `paragraph()`
|
|
353
|
-
// gives: a local copy is the defect, not a style choice.
|
|
354
|
-
// The closing-mark class `[*_)"'\u201d\u2019]*` appears once, not twice: it
|
|
355
|
-
// used to be spelled again inside the abbreviation guard's own lookbehind,
|
|
356
|
-
// two copies of the same set free to drift apart on a future edit. Nesting
|
|
357
|
-
// the abbreviation check onto the `.` alternative alone (never `!`/`?`, which
|
|
358
|
-
// no abbreviation ends in) keeps the same splits with one copy (verified:
|
|
359
|
-
// 20000 random terminator/closer/abbreviation fixtures and every .md/.mjs/
|
|
360
|
-
// .yml/.sh file in this repo split identically both ways).
|
|
361
|
-
//
|
|
362
|
-
// A footnote-style marker straight behind the terminator — `.[1]`, `?[^2]`,
|
|
363
|
-
// `.[1][2]`, closing marks allowed after it — is part of the sentence it
|
|
364
|
-
// annotates, and the end falls after it (#1987). Without it a `.[1]` ended
|
|
365
|
-
// nothing, so two real sentences merged: the silent direction for a pin that
|
|
366
|
-
// needs a word IN its sentence, which could borrow it from the next one. The
|
|
367
|
-
// marker holds at least one character and no whitespace, and must follow a
|
|
368
|
-
// terminator that closes a word, a digit, a code span, a paren or a curly
|
|
369
|
-
// quote. A word here is a letter in any script, combining marks on it
|
|
370
|
-
// included (#2041): an NFD "é" is `e` + U+0301 and Devanagari "हिंदी" ends in
|
|
371
|
-
// the vowel sign U+0940, so a word's last code point can be a mark, and the
|
|
372
|
-
// guard reads past any run of them to whichever of the five bases above sits
|
|
373
|
-
// under it. A mark with no accepted base under it closes nothing. Code quoted
|
|
374
|
-
// in prose puts brackets straight after a period too, with whitespace after
|
|
375
|
-
// them — jq's
|
|
376
|
-
// `'.[] | …'`, JavaScript's `m?.[1] || …`,
|
|
377
|
-
// a glob's `*.[0-9]*' ` — behind a quote, a `?`, a `*` or a space, and jq's
|
|
378
|
-
// `.jobs.[] | …` or `.jobs.[0, 1] | …` behind a word, with an empty or
|
|
379
|
-
// comma-spaced bracket. Measured over every tracked .md/.mjs/.yml/.sh/.json
|
|
380
|
-
// file: a bare `\[[^\]]*\]` after the terminator split 15 such code spots,
|
|
381
|
-
// and this rule none — its one new split is a real footnote, `only.[^1] `,
|
|
382
|
-
// in a citation-sweep fixture. Three shapes are known and left unfixed,
|
|
383
|
-
// none of them occurring in that same corpus today: a marker behind bold
|
|
384
|
-
// or italics (`**late**.[1]`) still merges, same as before; so does one
|
|
385
|
-
// behind a closing quote or paren that itself follows the terminator
|
|
386
|
-
// (`"stop."[1]`, `(noon.)[1]`) — the guard only tolerates closing marks
|
|
387
|
-
// AFTER the bracket run, not before it; and a single, unspaced jq index
|
|
388
|
-
// straight behind a word (`.jobs.[0] | …`) is the same shape as a real
|
|
389
|
-
// marker at the regex level, so it still splits — wrongly, the direction
|
|
390
|
-
// this rule exists to avoid.
|
|
391
|
-
export const sentences = (block) =>
|
|
392
|
-
block.split(/(?<=(?:[!?]|(?<!(?<![A-Za-z\d])(?:e\.g|i\.e|cf|viz|vs))\.)(?:(?<=[\p{L}\p{N}`)\u201d\u2019]\p{M}*[.!?])(?:\[[^\]\s]+\])+)?[*_)"'\u201d\u2019]*)\s+/u);
|
|
393
|
-
|
|
394
|
-
// A phrase-bounded slice — `paragraph`'s shape, with the end bound a second
|
|
395
|
-
// phrase instead of the next blank line. `from` is `anchorAt`'s own
|
|
396
|
-
// exactly-once anchor; `to` is a phrase asserted exactly once inside the
|
|
397
|
-
// optional `bound` — a regex searched for in the text after `from`, so the
|
|
398
|
-
// search for `to` stops there rather than reaching the end of the document.
|
|
399
|
-
// Without `bound`, `to` is searched over the rest of the document, the same
|
|
400
|
-
// reach as an unbounded `between`.
|
|
401
|
-
//
|
|
402
|
-
// `bound` exists for the reason `between`'s own header gives for its literal
|
|
403
|
-
// end anchor, applied to a phrase search instead of `indexOf`: an unbounded
|
|
404
|
-
// search for `to` is satisfiable from a LATER, unrelated occurrence once the
|
|
405
|
-
// true `to` is reworded away, silently widening the slice into whatever
|
|
406
|
-
// follows. A `bound` that excludes any such later occurrence turns that
|
|
407
|
-
// silent widening into a throw: the exactly-once assert below then sees
|
|
408
|
-
// zero hits inside the bound, never one from past it.
|
|
409
|
-
//
|
|
410
|
-
// `to`'s "occurs more than once" assert is scoped to `bound` too, not the
|
|
411
|
-
// whole rest of the document — a restatement of `to` PAST the bound must
|
|
412
|
-
// not count, or a pin could never pass once any later, unrelated copy of
|
|
413
|
-
// the same phrase exists anywhere below it.
|
|
414
|
-
//
|
|
415
|
-
// A bound can also match BEFORE an intact `to` — e.g. a reflow that puts
|
|
416
|
-
// `3.` at column 0 mid-step, which the `\n\d+\.\s` list-item bound reads as the
|
|
417
|
-
// next item, and CommonMark renders as one too, so the red is real. Reporting
|
|
418
|
-
// that as `to` having "moved" sends the reader after an edit to an anchor
|
|
419
|
-
// nobody touched, so a zero-hit scope looks past the cut before choosing the
|
|
420
|
-
// message: a `to` still found there names the line the bound first matched
|
|
421
|
-
// instead. That also covers a `to` reworded away with a later copy left past
|
|
422
|
-
// the bound — the helper cannot tell the two apart, so the message names both.
|
|
423
|
-
// A `to` found nowhere after `from` keeps "moved". Every one of these throws;
|
|
424
|
-
// only which message a red carries changes.
|
|
425
|
-
//
|
|
426
|
-
// The line named is the one the bound's match starts on, past one leading
|
|
427
|
-
// `\n` the match itself consumed — that newline terminates the text before
|
|
428
|
-
// the match, which is how a bound says "at the start of a line". Only one: a
|
|
429
|
-
// blank-line bound (`\n\n`, `\n\s*\n`, a zero-width `^$` with `m`) names the
|
|
430
|
-
// blank line it matched, never the next paragraph's text past it (#2268).
|
|
431
|
-
//
|
|
432
|
-
// Extracted (#1611) from `finisher-dispatch-premise-prose.test.mjs`'s
|
|
433
|
-
// `dispatchPremise`, which hand-rolled this exact shape locally — `paragraph`'s
|
|
434
|
-
// own header above names the class: a local copy is the defect, not a style
|
|
435
|
-
// choice.
|
|
436
|
-
export function betweenPhrases(text, from, to, what, { bound } = {}) {
|
|
437
|
-
const at = anchorAt(text, from, what);
|
|
438
|
-
const rest = text.slice(at);
|
|
439
|
-
const toPhrase = phrase(to);
|
|
440
|
-
let scope = rest;
|
|
441
|
-
let boundEnd = -1;
|
|
442
|
-
let boundLead = 0;
|
|
443
|
-
if (bound) {
|
|
444
|
-
// `exec` rather than `search` for the match text too; a copy without `g`
|
|
445
|
-
// so it starts at index 0 the way `search` does.
|
|
446
|
-
const boundMatch = new RegExp(bound, bound.flags.replace("g", "")).exec(rest);
|
|
447
|
-
boundEnd = boundMatch ? boundMatch.index : -1;
|
|
448
|
-
boundLead = boundMatch?.[0].startsWith("\n") ? 1 : 0;
|
|
449
|
-
assert.notEqual(
|
|
450
|
-
boundEnd,
|
|
451
|
-
-1,
|
|
452
|
-
`${what}: end bound no longer matches after "${from}" — re-anchor this test, never widen it to the whole document`,
|
|
453
|
-
);
|
|
454
|
-
scope = rest.slice(0, boundEnd);
|
|
455
|
-
}
|
|
456
|
-
const hits = [...scope.matchAll(new RegExp(toPhrase.source, "g"))];
|
|
457
|
-
// Zero hits inside the bound but one in `rest` means that hit ends past the
|
|
458
|
-
// cut — a hit ending inside it would have matched `scope` — so this also
|
|
459
|
-
// catches a bound that cuts through the middle of `to`.
|
|
460
|
-
if (hits.length === 0 && bound && toPhrase.test(rest)) {
|
|
461
|
-
const cut = at + boundEnd + boundLead;
|
|
462
|
-
const lineStart = text.lastIndexOf("\n", cut - 1) + 1;
|
|
463
|
-
const lineEnd = text.indexOf("\n", cut);
|
|
464
|
-
const line = text.slice(lineStart, lineEnd === -1 ? text.length : lineEnd);
|
|
465
|
-
const excerpt = line.length > 72 ? `${line.slice(0, 72)}…` : line;
|
|
466
|
-
const lineNo = text.slice(0, lineStart).split("\n").length;
|
|
467
|
-
assert.fail(
|
|
468
|
-
`${what}: slice end anchor "${to}" lies past the end bound, which first matched at line ${lineNo} ("${excerpt}") — an edit put a block boundary inside the slice, or reworded the anchor away and left a later copy past the bound; fix whichever it is, never widen the bound`,
|
|
469
|
-
);
|
|
470
|
-
}
|
|
471
|
-
assert.notEqual(
|
|
472
|
-
hits.length,
|
|
473
|
-
0,
|
|
474
|
-
`${what}: slice end anchor "${to}" moved — re-anchor this test, never widen it to the whole file`,
|
|
475
|
-
);
|
|
476
|
-
assert.equal(
|
|
477
|
-
hits.length,
|
|
478
|
-
1,
|
|
479
|
-
bound
|
|
480
|
-
? `${what}: slice end anchor "${to}" occurs ${hits.length} times inside the bound — a pin would bind the wrong copy; narrow the anchor`
|
|
481
|
-
: `${what}: slice end anchor "${to}" occurs ${hits.length} times — a pin would bind the wrong copy; narrow the anchor`,
|
|
482
|
-
);
|
|
483
|
-
return scope.slice(0, hits[0].index);
|
|
484
|
-
}
|
|
485
|
-
|
|
486
|
-
// Anchor matching only — never the compared text, so a caller that turns this
|
|
487
|
-
// on still returns raw bytes with emphasis intact. Exported so a pin comparing
|
|
488
|
-
// two files' copies of the same block (`**` moving on one side is real drift)
|
|
489
|
-
// does not hand-roll a second definition to locate the block it is comparing.
|
|
490
|
-
export const unemphasized = (s) => s.replace(/\*\*/g, "");
|
|
491
|
-
|
|
492
|
-
// Maps an offset found in `unemphasized(text)` back to `text`, stopping before
|
|
493
|
-
// any `**` at that point rather than after it — so a slice starting there
|
|
494
|
-
// keeps the marker instead of silently dropping it from the raw bytes a
|
|
495
|
-
// caller goes on to compare or substring-replace.
|
|
496
|
-
function rawOffset(text, unemphasizedOffset) {
|
|
497
|
-
let seen = 0;
|
|
498
|
-
let i = 0;
|
|
499
|
-
while (seen < unemphasizedOffset) {
|
|
500
|
-
if (text.startsWith("**", i)) i += 2;
|
|
501
|
-
else {
|
|
502
|
-
seen += 1;
|
|
503
|
-
i += 1;
|
|
504
|
-
}
|
|
505
|
-
}
|
|
506
|
-
return i;
|
|
507
|
-
}
|
|
508
|
-
|
|
509
|
-
// The offset of an anchor that must occur EXACTLY ONCE, asserted by count because
|
|
510
|
-
// a search takes the FIRST match silently, so an un-gutter'd restatement of the
|
|
511
|
-
// anchored block ABOVE the real one — the shape this repo's prose already uses
|
|
512
|
-
// where a phase quotes a member prompt back at itself — binds the pin to the copy
|
|
513
|
-
// while the real rule is gutted. A blockquoted copy is harmless (`\s+` cannot
|
|
514
|
-
// span the `>` gutter); a plain one is not. A missing anchor throws rather than
|
|
515
|
-
// widening, for `paragraph`'s reason.
|
|
516
|
-
//
|
|
517
|
-
// Split out of `paragraph` rather than left inside it because a slice whose END
|
|
518
|
-
// bound is not a blank line needs the same guarantee and must not copy it:
|
|
519
|
-
// `runAbove` below anchors on a declaration and takes the comment block ABOVE
|
|
520
|
-
// it, bounded by code at both ends, and imports this half rather than
|
|
521
|
-
// re-deriving it.
|
|
522
|
-
// `emphasisTolerant` is opt-in and off by default: most anchors here are
|
|
523
|
-
// typed WITH `**` when the source is (see e.g. `ancestry-check-position-prose
|
|
524
|
-
// .test.mjs`), so matching against `unemphasized(text)` unconditionally would
|
|
525
|
-
// break every anchor that itself carries emphasis. On, it lets an anchor
|
|
526
|
-
// typed WITHOUT `**` still find its target after a meaning-preserving
|
|
527
|
-
// emphasis move on the matched words — measured need in
|
|
528
|
-
// `tracker-block-copy-prose.test.mjs`, where the opener anchor has none.
|
|
529
|
-
export function anchorAt(text, anchor, what, { emphasisTolerant = false } = {}) {
|
|
530
|
-
const haystack = emphasisTolerant ? unemphasized(text) : text;
|
|
531
|
-
const hits = [...haystack.matchAll(new RegExp(phrase(anchor).source, "g"))];
|
|
532
|
-
assert.notEqual(hits.length, 0, `${what}: slice anchor "${anchor}" moved — re-anchor this test, never widen it to the whole file`);
|
|
533
|
-
assert.equal(hits.length, 1, `${what}: slice anchor "${anchor}" occurs ${hits.length} times — a pin would bind the wrong copy; narrow the anchor`);
|
|
534
|
-
return emphasisTolerant ? rawOffset(text, hits[0].index) : hits[0].index;
|
|
535
|
-
}
|
|
536
|
-
|
|
537
|
-
// The comment run immediately ABOVE a declaration, gutter intact, bounded by
|
|
538
|
-
// CODE at both ends. The bound `paragraph` cannot give a SOURCE file: there the
|
|
539
|
-
// blank line after a comment block sits below the code the block documents, so
|
|
540
|
-
// a blank-line slice runs past the prose into live code and a decoy pasted
|
|
541
|
-
// there takes the pin green (measured #695). Anchoring on the declaration
|
|
542
|
-
// instead puts code at both ends, so no edit to the prose the slice holds can
|
|
543
|
-
// move either bound — which is the whole reason a pin over source prose gets to
|
|
544
|
-
// be a slice at all rather than a search of the file.
|
|
545
|
-
//
|
|
546
|
-
// `marker` is the gutter literal (`#` for shell, `//` for `.mjs`), matched only
|
|
547
|
-
// as a line's FIRST token after indent: a `//` inside a string on a code line
|
|
548
|
-
// does not extend the run, which is what keeps "bounded by code" true rather
|
|
549
|
-
// than merely stated. It is escaped, so a `*` continuation gutter cannot turn
|
|
550
|
-
// the run into `.*` and swallow every line above the anchor. No
|
|
551
|
-
// `emphasisTolerant` pass-through, unlike `paragraph`: `**` is markdown
|
|
552
|
-
// emphasis, and every anchor this bound takes is a declaration in source.
|
|
553
|
-
//
|
|
554
|
-
// Returns "" rather than throwing when nothing but code sits above the anchor,
|
|
555
|
-
// because its callers diagnose that differently and both are right:
|
|
556
|
-
// `gp-sep-invariant-prose.test.mjs` slices at module load and wants the empty
|
|
557
|
-
// case as a NAMED test rather than an import-time crash, while
|
|
558
|
-
// `ed9f2967^:plugin/scripts/review-pr-inbound-citation-prose.test.mjs` asserted
|
|
559
|
-
// non-empty on the spot.
|
|
560
|
-
// The exactly-once half is not the caller's and is not optional: it is
|
|
561
|
-
// `anchorAt`'s, so a second copy of the anchored declaration reds here instead
|
|
562
|
-
// of silently binding the pin to whichever copy comes first.
|
|
563
|
-
//
|
|
564
|
-
// The one thing this closes that all three hand-rolled copies had open: an
|
|
565
|
-
// INDENTED anchor. Each copy ended its run at `$`, so a declaration inside a
|
|
566
|
-
// function body — the text above it ending in the anchor's own indent, not a
|
|
567
|
-
// newline — matched nothing and handed back "" with the comment sitting right
|
|
568
|
-
// there, which reads as "the block is gone" and names the wrong fault. The
|
|
569
|
-
// tolerated gap is `[ \t]*` and nothing else, so it cannot widen: no code can
|
|
570
|
-
// hide in horizontal whitespace, the run itself is still returned without that
|
|
571
|
-
// indent, and all three migrated anchors sit at column 0 today, so the bytes
|
|
572
|
-
// they get back are unchanged.
|
|
573
|
-
//
|
|
574
|
-
// #1622: a SAME-LINE keyword prefix needed the same fix as the indent did.
|
|
575
|
-
// `export const TARGET` anchored on `const TARGET` left `"export "` — not
|
|
576
|
-
// whitespace — at the tail, so `[ \t]*$` could not match either, and the run
|
|
577
|
-
// above came back "" while the comment sat right there, the identical wrong
|
|
578
|
-
// fault the indent fix closed. The gap this bound now also tolerates is
|
|
579
|
-
// `DECLARATION_PREFIX_KEYWORDS`, a fixed, closed list of the JS/TS keywords
|
|
580
|
-
// that legally sit between a comment block and the declaration keyword an
|
|
581
|
-
// anchor starts with (`export`, `default`, `async`, visibility and
|
|
582
|
-
// modifier keywords, …) — never arbitrary text. Each tolerated token must
|
|
583
|
-
// still be followed by more whitespace before the next token or the anchor,
|
|
584
|
-
// so the widening cannot reach past the anchor's own line: a decoy code line
|
|
585
|
-
// sitting between the comment block and the anchor still breaks the run,
|
|
586
|
-
// because that line does not end in `\n` followed only by keywords and
|
|
587
|
-
// whitespace up to the anchor.
|
|
588
|
-
//
|
|
589
|
-
// The single definition of this bound in this directory (#1604), for
|
|
590
|
-
// `paragraph`'s reason and after its history: three pins hand-rolled this same
|
|
591
|
-
// regex locally first, each carrying its own chance of dropping a guard. A pin
|
|
592
|
-
// needing this bound imports it — a local copy is the defect, not a style
|
|
593
|
-
// choice.
|
|
594
|
-
|
|
595
|
-
// The closed list `runAbove`'s tail bound tolerates between the comment block
|
|
596
|
-
// and the anchor's own declaration keyword. Deliberately narrow: modifiers
|
|
597
|
-
// and the declaration keywords themselves, nothing that could also open a
|
|
598
|
-
// statement a comment block should not be glued to (no `if`, `for`, `return`,
|
|
599
|
-
// …). Case-sensitive and matched as whole identifier tokens only (surrounded
|
|
600
|
-
// by mandatory whitespace), so it cannot partially match inside a longer
|
|
601
|
-
// identifier.
|
|
602
|
-
const DECLARATION_PREFIX_KEYWORDS =
|
|
603
|
-
"export|default|async|function|class|const|let|var|public|private|protected|static|readonly|abstract|declare|override|get|set";
|
|
604
|
-
|
|
605
|
-
export function runAbove(text, anchor, what, marker) {
|
|
606
|
-
assert.ok(marker, `${what}: runAbove needs a non-empty marker — empty matches every line above the anchor instead of just the comment block`);
|
|
607
|
-
const above = text.slice(0, anchorAt(text, anchor, what));
|
|
608
|
-
const run = above.match(
|
|
609
|
-
new RegExp(`(?:^|\\n)((?:[ \\t]*${escapeRe(marker)}[^\\n]*\\n)+)[ \\t]*(?:(?:${DECLARATION_PREFIX_KEYWORDS})[ \\t]+)*$`),
|
|
610
|
-
);
|
|
611
|
-
return run ? run[1] : "";
|
|
612
|
-
}
|
|
613
|
-
|
|
614
|
-
// A shell comment block wraps at `#`, so a pinned phrase can break across lines
|
|
615
|
-
// with the comment gutter, not whitespace, at the break — `\s+` does not span a
|
|
616
|
-
// `#`. Strip the gutter and rejoin with the single inter-word space a wrap point
|
|
617
|
-
// replaces.
|
|
618
|
-
export const stripHashGutter = (text) => text.split("\n").map((l) => l.replace(/^\s*#\s?/, "")).join(" ");
|
|
619
|
-
|
|
620
|
-
// A markdown blockquote wraps every line in a leading `>` gutter, so a
|
|
621
|
-
// dispatch instruction embedded inside a quoted prompt (#1341's grandchild
|
|
622
|
-
// recipe, nested under run-team's fix-applier prompt) does not begin with its
|
|
623
|
-
// harness token until the gutter is stripped. Mirrors stripHashGutter's
|
|
624
|
-
// reason for existing, one gutter shape later.
|
|
625
|
-
export const stripQuoteGutter = (text) => text.split("\n").map((l) => l.replace(/^\s*>+\s?/, "")).join("\n");
|
|
626
|
-
|
|
627
|
-
// One verbatim dispatch block, bounded by its OWN quote run rather than by the
|
|
628
|
-
// next block's opening words, and returned with the gutter intact so the caller
|
|
629
|
-
// owns the normalization. #1002: a whole-block golden fixture needs a slice that
|
|
630
|
-
// is exactly one block, and the neighbour-anchored form every presence pin in
|
|
631
|
-
// this directory uses is not one. Measured on `run-team/SKILL.md`:
|
|
632
|
-
// `dispatch-block-pins-prose.test.mjs`'s `commitBlock` runs from `Commit
|
|
633
|
-
// incrementally` to `**Every scratch file` and so already contains the whole
|
|
634
|
-
// stash-prohibition block that landed between them, and its `worktreeBlock`
|
|
635
|
-
// likewise contains the distilled-brief block. Both slices silently grew a
|
|
636
|
-
// second block, which a presence pin never notices and a golden could not
|
|
637
|
-
// survive: the fixture would be "this block plus whatever lands after it", and
|
|
638
|
-
// re-blessing it would bless the insertion too.
|
|
639
|
-
//
|
|
640
|
-
// A maximal run of `>` lines is the bound markdown itself uses, so it cannot
|
|
641
|
-
// drift: a sentence appended inside the block stays inside the run, and a block
|
|
642
|
-
// spliced in after it is a SEPARATE run — which is what makes "every run in the
|
|
643
|
-
// region has a fixture" a check a caller can make at all.
|
|
644
|
-
//
|
|
645
|
-
// Column-0 `>` only, matching how every pin in this directory already strips
|
|
646
|
-
// this gutter (`/^>\s?/`) and how `member-prompt-prose.test.mjs` asserts every
|
|
647
|
-
// member-facing line carries one. An indented gutter is not silently absorbed
|
|
648
|
-
// here: it ends the run, so the block reds against its fixture instead of
|
|
649
|
-
// passing under a changed markdown context.
|
|
650
|
-
export function quoteBlocks(text) {
|
|
651
|
-
const runs = [];
|
|
652
|
-
let open = null;
|
|
653
|
-
for (const line of text.split("\n")) {
|
|
654
|
-
if (line.startsWith(">")) (open ??= []).push(line);
|
|
655
|
-
else if (open) {
|
|
656
|
-
runs.push(open.join("\n"));
|
|
657
|
-
open = null;
|
|
658
|
-
}
|
|
659
|
-
}
|
|
660
|
-
if (open) runs.push(open.join("\n"));
|
|
661
|
-
return runs;
|
|
662
|
-
}
|
|
663
|
-
|
|
664
|
-
// The one quote run that BEGINS with `opener`, for a caller comparing that run
|
|
665
|
-
// against a golden copy. Both failure directions throw rather than return, for
|
|
666
|
-
// the reason a golden makes sharper than a presence pin does: an extractor that
|
|
667
|
-
// returns "" for a block it cannot find compares empty against empty — or
|
|
668
|
-
// against a fixture nobody re-blessed — and a deleted block passes.
|
|
669
|
-
//
|
|
670
|
-
// Anchored at the run's START, not anywhere inside it, so the identifier names
|
|
671
|
-
// the block rather than merely occurring in it: a later block quoting the same
|
|
672
|
-
// opening words cannot answer to it. Matched through `phrase()` on the
|
|
673
|
-
// flattened run for `paragraph`'s reason — a literal `indexOf` reds on a rewrap
|
|
674
|
-
// that the opener itself survives, turning a reflow into a red.
|
|
675
|
-
// `emphasisTolerant` mirrors `anchorAt`'s option and for the same reason: off
|
|
676
|
-
// by default because an opener typed WITH `**` must still match one, on so an
|
|
677
|
-
// opener typed WITHOUT it can find a target whose emphasis moved. Safe to
|
|
678
|
-
// apply only to the match, never to `hits[0]` itself — the returned run keeps
|
|
679
|
-
// its `**` intact, since that is the raw byte content callers go on to
|
|
680
|
-
// compare or rewrap.
|
|
681
|
-
export function quoteBlock(text, opener, what, { emphasisTolerant = false } = {}) {
|
|
682
|
-
const head = new RegExp(`^${phrase(opener).source}`);
|
|
683
|
-
const view = emphasisTolerant ? unemphasized : (s) => s;
|
|
684
|
-
const hits = quoteBlocks(text).filter((run) =>
|
|
685
|
-
head.test(view(run.split("\n").map((l) => l.replace(/^>[ \t]?/, "")).join(" ").split(/\s+/).join(" ").trim())),
|
|
686
|
-
);
|
|
687
|
-
assert.notEqual(
|
|
688
|
-
hits.length,
|
|
689
|
-
0,
|
|
690
|
-
`${what}: no quote block opens on "${opener}" — the block was deleted, un-quoted, or its opening words changed. Re-anchor or restore it; this throws rather than comparing a golden fixture against nothing`,
|
|
691
|
-
);
|
|
692
|
-
assert.equal(
|
|
693
|
-
hits.length,
|
|
694
|
-
1,
|
|
695
|
-
`${what}: ${hits.length} quote blocks open on "${opener}" — a fixture would bind to whichever came first; narrow the opener`,
|
|
696
|
-
);
|
|
697
|
-
return hits[0];
|
|
698
|
-
}
|
|
699
|
-
|
|
700
|
-
// A `.mjs` header comment's prose is wrapped at `// `, one gutter per
|
|
701
|
-
// physical line, so a scan that wants the prose as CONTINUOUS text — not
|
|
702
|
-
// broken by a comment marker on every line — has to strip it first. Only
|
|
703
|
-
// whole-line `//` comments are stripped (mirrors `stripComments.mjs`'s own
|
|
704
|
-
// line-based-on-purpose note) — a trailing `code; // a note` is not a
|
|
705
|
-
// comment LINE, it is code with a trailing note, and stripping mid-line
|
|
706
|
-
// would blur that distinction and corrupt the code beside it.
|
|
707
|
-
export const stripSlashGutter = (text) => text.split("\n").map((l) => l.replace(/^\s*\/\/\s?/, "")).join("\n");
|
|
708
|
-
|
|
709
|
-
// A paragraph's physical line breaks, removed. `phrase()` holds this axis for
|
|
710
|
-
// a pin that KNOWS the words it is looking for; this holds it for a pin that
|
|
711
|
-
// DERIVES its subjects from the document. Those are different problems: a
|
|
712
|
-
// literal you can pre-tolerate with `\s+`, but a scan like `see \*\*([^*\n]+)\*\*`
|
|
713
|
-
// has to keep `\n` out of its own character classes to stay one clause, and a
|
|
714
|
-
// generative regex cannot be made whitespace-tolerant and bounded at once.
|
|
715
|
-
// #1609: rewrapping `review-and-fix.md`'s step 4 — pure reflow, not one word
|
|
716
|
-
// changed — took that scan from 5 pointer names to 3, and the pins for the two
|
|
717
|
-
// it lost stopped existing rather than failing. Three separate breaks, all the
|
|
718
|
-
// same shape: the literal space in `see `, the literal space in the list
|
|
719
|
-
// separator, and `\n` excluded from the name class. Tolerating them one at a
|
|
720
|
-
// time in the regex is how the bound gets lost — move the WRAP, not the scan.
|
|
721
|
-
//
|
|
722
|
-
// The view, never the file: this returns a copy to scan, and `lineAt` maps a
|
|
723
|
-
// hit in it back to the physical line a reader has to open. A pin reporting an
|
|
724
|
-
// offset into the joined copy names a line that does not exist in the source.
|
|
725
|
-
//
|
|
726
|
-
// Joining only ADDS matches, so a pin cannot lose a hit to this — but `^` moves,
|
|
727
|
-
// and that is the point in both directions. A `**Bold**` that a wrap happens to
|
|
728
|
-
// push to column 0 mid-paragraph reads as block-leading to a `/^\*\*/m` scan
|
|
729
|
-
// today, which is the mirror false GREEN: a pointer resolving to a target the
|
|
730
|
-
// reflow invented. On the joined view it is mid-line again, where it belongs.
|
|
731
|
-
//
|
|
732
|
-
// What is NOT modelled: an indented (4-space) code block. Its second and later
|
|
733
|
-
// lines are indistinguishable from a deeply-indented list continuation without
|
|
734
|
-
// a real block parser, so they join and the block collapses to one line. Fenced
|
|
735
|
-
// code is modelled and never joined — scan code through a fence, not an indent.
|
|
736
|
-
|
|
737
|
-
// Starts a new block, so it never continues its predecessor.
|
|
738
|
-
const BLOCK_START = /^ {0,3}(?:#{1,6}\s|[-*+](?:\s|$)|\d+[.)]\s|>|\||```|~~~|(?:-{3,}|\*{3,}|_{3,})\s*$)/;
|
|
739
|
-
// Complete in itself, so nothing continues IT. Narrower than
|
|
740
|
-
// BLOCK_START on purpose: a list item and a blockquote both take lazy
|
|
741
|
-
// continuation lines, a heading and a thematic break take none.
|
|
742
|
-
const BLOCK_END = /^ {0,3}(?:#{1,6}\s|\||```|~~~|(?:-{3,}|\*{3,}|_{3,})\s*$)/;
|
|
743
|
-
// Two trailing SPACES or a trailing backslash is markdown's hard line break —
|
|
744
|
-
// an authored break, not a wrap point, and removing it would change the
|
|
745
|
-
// render. Spaces only, per CommonMark: a tab does not make one, and widening
|
|
746
|
-
// this to `[ \t]{2}` refuses to join a line an editor left ending in a
|
|
747
|
-
// space and a tab, which is the same false red one layer down.
|
|
748
|
-
const HARD_BREAK = /(?: {2}|\\)$/;
|
|
749
|
-
|
|
750
|
-
export function logicalLines(text) {
|
|
751
|
-
const src = text.split("\n");
|
|
752
|
-
const joins = new Array(src.length).fill(false);
|
|
753
|
-
let fenced = false;
|
|
754
|
-
for (let i = 0; i < src.length; i++) {
|
|
755
|
-
const fence = /^ {0,3}(?:```|~~~)/.test(src[i]);
|
|
756
|
-
joins[i] =
|
|
757
|
-
i > 0 &&
|
|
758
|
-
!fenced &&
|
|
759
|
-
!fence &&
|
|
760
|
-
/\S/.test(src[i - 1]) &&
|
|
761
|
-
/\S/.test(src[i]) &&
|
|
762
|
-
!HARD_BREAK.test(src[i - 1]) &&
|
|
763
|
-
!BLOCK_END.test(src[i - 1]) &&
|
|
764
|
-
!BLOCK_START.test(src[i]);
|
|
765
|
-
if (fence) fenced = !fenced;
|
|
766
|
-
}
|
|
767
|
-
|
|
768
|
-
const starts = new Array(src.length);
|
|
769
|
-
const out = [];
|
|
770
|
-
let offset = 0;
|
|
771
|
-
for (let i = 0; i < src.length; i++) {
|
|
772
|
-
// A wrap point is one inter-word space, so the indent the wrap introduced
|
|
773
|
-
// comes off and exactly one space goes back — `stripHashGutter`'s rejoin,
|
|
774
|
-
// one gutter shape fewer. The trailing trim is why that space is never two.
|
|
775
|
-
let line = joins[i] ? src[i].replace(/^[ \t]+/, "") : src[i];
|
|
776
|
-
if (joins[i + 1]) line = line.replace(/[ \t]+$/, "");
|
|
777
|
-
if (i > 0) {
|
|
778
|
-
const sep = joins[i] ? " " : "\n";
|
|
779
|
-
out.push(sep);
|
|
780
|
-
offset += sep.length;
|
|
781
|
-
}
|
|
782
|
-
starts[i] = offset;
|
|
783
|
-
out.push(line);
|
|
784
|
-
offset += line.length;
|
|
785
|
-
}
|
|
786
|
-
|
|
787
|
-
// Strictly increasing — an unjoined line contributes its own `\n` and a
|
|
788
|
-
// joined one is non-blank by construction — so this bisect has no ties to
|
|
789
|
-
// break and an offset inside line i answers i, never i+1.
|
|
790
|
-
const joined = out.join("");
|
|
791
|
-
const lineAt = (at) => {
|
|
792
|
-
assert.equal(
|
|
793
|
-
at >= 0 && at < joined.length,
|
|
794
|
-
true,
|
|
795
|
-
`logicalLines().lineAt(${at}) out of range: the joined view is ${joined.length} bytes — a valid offset comes from a match against \`text\`, never a caller's own guess`,
|
|
796
|
-
);
|
|
797
|
-
let lo = 0;
|
|
798
|
-
let hi = starts.length - 1;
|
|
799
|
-
while (lo < hi) {
|
|
800
|
-
const mid = (lo + hi + 1) >> 1;
|
|
801
|
-
if (starts[mid] <= at) lo = mid;
|
|
802
|
-
else hi = mid - 1;
|
|
803
|
-
}
|
|
804
|
-
return lo + 1;
|
|
805
|
-
};
|
|
806
|
-
return { text: joined, lineAt };
|
|
807
|
-
}
|