@vincemakes/kiso-tools-node 0.40.2 → 0.40.3

package/dist/index.js

@@ -31,7 +31,7 @@ import { strippedShellEnv } from "./secret-env.js";
 import { contentRevision, normalizeRevision, postEffectEscape, precondition, publishNewFile, revalidateBeforeRename } from "./wr1.js";
 import { isProtectedPath, protectedIdentity, protectedRefusalText, readUnlessProtected } from "./protected.js";
 import { CORPUS_MAX_DEPTH, globToRegExp, walkCorpus } from "./corpus.js";
-import { describeSearchMiss } from "./search-miss.js";
+import { describeSearchMiss, regionForSearch } from "./search-miss.js";
 /**
  * TUI2-R1 (C) — THE SHELL PROGRESS SIDECAR.
  *
@@ -979,10 +979,49 @@ function describeMatchLines(text, offsets, exhaustive) {
     const tail = !exhaustive ? " and more" : hidden > 0 ? ` and ${hidden} more` : "";
     return `${shown.length === 1 && tail === "" ? "line" : "lines"} ${shown.join(", ")}${tail}`;
 }
+/**
+ * ACI-3 — a stale citation's refusal, made retry-ready: where each hunk
+ * stands in the file AS IT IS, applied in order the way the edit would
+ * apply them. On the owner's disk 102 refusals were "changed since": 41
+ * cited a revision older than one they had already been shown, 40 cited
+ * one no tool had returned in the session, 21 met a real outside change.
+ * Every one was followed by a read. A hunk that still matches once needs
+ * nothing more than the new revision; one that does not gets the text
+ * where it should land. At most three regions: the rest are named.
+ */
+function staleReport(text, hunks, single) {
+    const out = ["the file as it is now:"];
+    let running = text;
+    let regions = 0;
+    for (let i = 0; i < hunks.length; i += 1) {
+        const h = hunks[i];
+        const name = single ? "the search" : `hunk ${i + 1}`;
+        const { count, offsets } = occurrencesOf(running, h.search);
+        if (count === 1) {
+            out.push(`  ${name}: matches once, ${describeMatchLines(running, offsets, true)}`);
+            running = running.slice(0, offsets[0]) + h.replace + running.slice(offsets[0] + h.search.length);
+            continue;
+        }
+        if (count > 1) {
+            out.push(`  ${name}: matches ${count} places (${describeMatchLines(running, offsets, count <= ACI2_OFFSETS_KEPT)}) — make it unique`);
+            continue;
+        }
+        const miss = describeSearchMiss(running, h.search);
+        out.push(`  ${name}: not found${miss ? `\n${miss.replace(/^/gm, "  ")}` : ""}`);
+        if (regions >= 3)
+            continue;
+        const region = regionForSearch(running, h.search);
+        if (region === null)
+            continue;
+        regions += 1;
+        out.push(`    lines ${region.from}–${region.to}:\n${region.body}`);
+    }
+    return out.join("\n");
+}
 export function editFileTool(opts) {
     return defineTool({
         name: "edit_file",
-        description: "Edit a workspace file at its latest revision (expectedRevision). ONE of: search+replace (must match exactly once), or edits (1-32 disjoint hunks resolved against the same snapshot, applied atomically).",
+        description: "Edit a file at its latest revision (expectedRevision): search+replace (must match once) OR edits (1-32 hunks, applied in order, all or nothing). A refusal shows the current text and ends on its [rev:X].",
         parameters: {
             type: "object",
             properties: {
@@ -991,10 +1030,13 @@ export function editFileTool(opts) {
                 replace: { type: "string", description: "Replacement text (single-hunk form)" },
                 edits: {
                     type: "array",
-                    description: "Batch form: 1-32 {search, replace} hunks, every search resolved against the SAME snapshot",
+                    description: "Batch form: 1-32 {search, replace} hunks, applied in order",
                     items: {
                         type: "object",
-                        properties: { search: { type: "string" }, replace: { type: "string" } },
+                        // ACI-3: `expectedRevision` inside a hunk is tolerated and ignored —
+                        // the call's own governs (5 refusals on the owner's disk, every
+                        // one with the top-level token present too).
+                        properties: { search: { type: "string" }, replace: { type: "string" }, expectedRevision: { type: "string" } },
                         required: ["search", "replace"],
                         additionalProperties: false,
                     },
@@ -1052,7 +1094,7 @@ export function editFileTool(opts) {
             let preservedMode;
             try {
                 if (!existsSync(full)) {
-                    return precondition(`edit_file: ${path} no longer exists — it was deleted since you read it`);
+                    return precondition(`edit_file: ${path} does not exist — to create it, use write_file with expectedRevision:"absent" (if you read it before, it was deleted since)`);
                 }
                 // WR-1: ONE byte snapshot feeds BOTH the validation and the
                 // mutation — no internal window between what was checked and
@@ -1076,31 +1118,37 @@ export function editFileTool(opts) {
                 if (bytes === null)
                     return precondition(protectedRefusalText("edit_file", path));
                 const current = contentRevision(bytes);
+                const text = bytes.toString("utf8");
                 if (current !== expectedRevision) {
-                    return precondition(`edit_file: ${path} changed since ${expectedRevision} — read it again and cite its [rev:…] line, then re-apply the change`);
+                    // ACI-3: the refusal carries the file as it is where each hunk
+                    // should land, and its current revision last — the retry needs
+                    // no read. The headline is unchanged (bench parsers read it).
+                    return precondition(`edit_file: ${path} changed since ${expectedRevision} — the file as it is now is below; re-apply the change against it, citing the [rev:…] on the last line\n${staleReport(text, hunks, edits === undefined)}\n[${current}]`);
                 }
-                const text = bytes.toString("utf8");
-                // WR-1E2: EVERY hunk resolves against THIS snapshot — never the
-                // output of an earlier hunk. All spans are known before any
-                // staging; overlaps refuse. Since ACI-2 a non-unique search is
-                // already refused above, so the overlap left to catch is two
-                // hunks aimed at the same unique text — never retargeted.
-                const spans = [];
+                // ACI-3 (superseding WR-1E2's one-snapshot rule): the hunks apply
+                // IN ORDER, each against the result of the ones before it — the
+                // way a person writes a second change after the first, and the
+                // reference implementation's multi-edit. 57 refusals on the
+                // owner's disk were "two hunks overlap": a second hunk written
+                // against the first one's output. Still all or nothing: nothing is
+                // written unless every hunk resolves, and each must match exactly
+                // once in the text it meets (ACI-2).
+                let edited = text;
                 for (let i = 0; i < hunks.length; i += 1) {
                     const h = hunks[i];
-                    const { count, offsets } = occurrencesOf(text, h.search);
+                    const { count, offsets } = occurrencesOf(edited, h.search);
+                    const which = hunks.length === 1 && edits === undefined ? "" : i === 0 ? " (hunk 1)" : ` (hunk ${i + 1}, after ${i === 1 ? "hunk 1" : `hunks 1–${i}`} applied)`;
                     if (count === 0) {
                         // WR-1A ④: the WORLD lacks the pattern (the input is
                         // fine) and nothing ran — precondition; the note never
-                        // rides an edit that wrote nothing.
-                        // The headline says WHAT failed; the detail says WHERE.
-                        // A refusal that names the divergence costs one line
-                        // here and saves a whole file read at the caller.
-                        const headline = hunks.length === 1 && edits === undefined
-                            ? `edit_file: pattern not found in ${path}`
-                            : `edit_file: pattern not found in ${path} (hunk ${i + 1})`;
-                        const detail = describeSearchMiss(text, h.search);
-                        return precondition(detail ? `${headline}\n${detail}` : headline);
+                        // rides an edit that wrote nothing. The headline says WHAT
+                        // failed; the detail says WHERE; ACI-3: the region is the
+                        // text the hunk met, and the file's revision ends it.
+                        const detail = describeSearchMiss(edited, h.search);
+                        const region = regionForSearch(edited, h.search);
+                        const seen = i === 0 ? "the file now has" : `the text hunk ${i + 1} met`;
+                        const block = region === null ? "" : `\n${seen}, lines ${region.from}–${region.to}:\n${region.body}`;
+                        return precondition(`edit_file: pattern not found in ${path}${which}${detail ? `\n${detail}` : ""}${block}\n[${current}]`);
                     }
                     // ACI-2: more than one resolution is a QUESTION, not an edit.
                     // Taking the first one wrote the wrong place and reported
@@ -1108,28 +1156,15 @@ export function editFileTool(opts) {
                     // have. The refusal carries the count and the lines so the
                     // call can be fixed without reading the file again.
                     if (count > 1) {
-                        const lines = describeMatchLines(text, offsets, count <= ACI2_OFFSETS_KEPT);
-                        const where = hunks.length === 1 && edits === undefined ? lines : `hunk ${i + 1}, ${lines}`;
-                        return precondition(`edit_file: pattern matches ${count} places in ${path} (${where}) — include enough surrounding text to make it unique`);
-                    }
-                    spans.push({ start: offsets[0], end: offsets[0] + h.search.length, replace: h.replace });
-                }
-                const bySpan = [...spans].sort((a, b) => a.start - b.start);
-                for (let i = 1; i < bySpan.length; i += 1) {
-                    if (bySpan[i].start < bySpan[i - 1].end) {
-                        return precondition(`edit_file: two hunks overlap in ${path} — make the searches disjoint`);
+                        const lines = describeMatchLines(edited, offsets, count <= ACI2_OFFSETS_KEPT);
+                        const where = which === "" ? lines : `${which.slice(2, -1)}, ${lines}`;
+                        return precondition(`edit_file: pattern matches ${count} places in ${path} (${where}) — include enough surrounding text to make it unique\n[${current}]`);
                     }
+                    edited = edited.slice(0, offsets[0]) + h.replace + edited.slice(offsets[0] + h.search.length);
                 }
                 // E group: safe replacement — never rewrite a shared external inode via a hard link.
                 // round 8: the edited file keeps its mode.
                 preservedMode = statSync(full).mode & 0o7777;
-                // The postimage: replacements applied highest-offset-first, so
-                // earlier spans never shift later coordinates — deterministic
-                // under any hunk ORDER (one observed state, one postimage).
-                let edited = text;
-                for (const sp of [...bySpan].reverse()) {
-                    edited = edited.slice(0, sp.start) + sp.replace + edited.slice(sp.end);
-                }
                 writeFileSync(tmp, edited, "utf8");
                 chmodSync(tmp, preservedMode);
                 // WR-1A ③: revalidate against the citation right before the

package/dist/search-miss.d.ts

@@ -22,3 +22,26 @@
  * to say. The caller owns the headline; this is what follows it.
  */
 export declare function describeSearchMiss(text: string, search: string): string;
+/**
+ * ACI-3 — the text a refused edit needs to be retried WITHOUT reading the
+ * file again. On the owner's disk every edit_file refusal was followed by
+ * a read before the retry: two extra requests per failure, 319 failures.
+ * The refusal already knows the file; showing the few lines where the
+ * search should land costs a couple of KB once instead of a whole read.
+ *
+ * Whole lines only (a cut mid-line is a lie about the file), `context`
+ * lines each side of the search's own span, cut at a line boundary under
+ * `maxChars`. Centred on the best partial match — the longest matching
+ * prefix at its earliest occurrence, the place describeSearchMiss names —
+ * and null when that prefix is too short to say where the caller aimed.
+ */
+export interface Region {
+    /** 1-based, inclusive. */
+    readonly from: number;
+    readonly to: number;
+    readonly body: string;
+}
+export declare function regionForSearch(text: string, search: string, context?: number, maxChars?: number): Region | null;
+/** The whole lines from `context` before `offset`'s line to `context` after
+ *  the `span` lines that start there. */
+export declare function regionAt(text: string, offset: number, span: number, context?: number, maxChars?: number): Region;

package/dist/search-miss.js

@@ -92,3 +92,32 @@ export function describeSearchMiss(text, search) {
         `  your search wanted: "${fragment(rest)}"`,
     ].join("\n");
 }
+export function regionForSearch(text, search, context = 6, maxChars = 1500) {
+    if (search.length === 0)
+        return null;
+    const matched = longestMatchingPrefix(text, search);
+    // A prefix of a character or two lands anywhere: its region would be
+    // noise that costs context. A quarter of the search (at most 16 chars)
+    // is where the match starts to say WHERE the caller was aiming.
+    if (matched === 0 || matched < Math.min(16, Math.ceil(search.length / 4)))
+        return null;
+    return regionAt(text, text.indexOf(search.slice(0, matched)), search.split("\n").length, context, maxChars);
+}
+/** The whole lines from `context` before `offset`'s line to `context` after
+ *  the `span` lines that start there. */
+export function regionAt(text, offset, span, context = 6, maxChars = 1500) {
+    const lines = text.split("\n");
+    const last = text.endsWith("\n") ? lines.length - 1 : lines.length;
+    const at = lineAt(text, offset);
+    const from = Math.max(1, at - context);
+    let to = Math.max(from, Math.min(last, at + Math.max(span, 1) - 1 + context));
+    let body = lines.slice(from - 1, to).join("\n");
+    while (body.length > maxChars && to > from) {
+        to -= 1;
+        body = lines.slice(from - 1, to).join("\n");
+    }
+    // one line longer than the cap on its own: the honest answer is the cut, said
+    if (body.length > maxChars)
+        body = `${body.slice(0, maxChars)}… (line ${from} continues)`;
+    return { from, to, body };
+}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vincemakes/kiso-tools-node",
-	"version": "0.40.2",
+	"version": "0.40.3",
 	"description": "kiso coding tools for Node hosts \u2014 read file, list directory, search text, write/edit file, shell command.",
 	"type": "module",
 	"license": "MIT",
@@ -25,7 +25,7 @@
 		"test": "vitest run"
 	},
 	"dependencies": {
-		"@vincemakes/kiso-core": "0.40.2",
+		"@vincemakes/kiso-core": "0.40.3",
 		"ignore": "^7.0.9"
 	},
 	"devDependencies": {