fix(search): a space after an operator's colon is part of the operator

CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 1.0.91
+
+- Fix: **a space after an operator's colon now works.** `title: ci time` was read as a bare `title:` with no value, reported as unreadable, and then searched as the two plain words `ci` and `time` — 113 sessions instead of the 7 that `title:ci time` returns. The space is what people type, and a bare `title:` meant nothing before, so for a plain word the rule only changes queries that were already being reported as errors. Applies to every operator (`pr:`, `has:`, `is:`, `after:`, `before:` and all six field terms), and a quoted phrase still works after the space (`title: "two words"`). **A token that is already a search term is never taken** — another operator or a PR reference — so `title: is:live` and `title: #137` keep searching for the live filter and for PR 137 respectively, each with its `title:` reported, rather than silently losing the term inside a field value. **Quotes are the escape hatch**, identically with and without the space: `title: "is:live"` and `title:"is:live"` both mean a title containing that text. When a taken value turns out to be unusable the warning now shows both halves (`ignored: after: soon`) instead of a bare `after:`
+
 ## 1.0.90
 
 - Fix: the **`match …` markers are outlined rather than filled.** They carried the same solid amber as the highlighted words, which was right while a row had at most one of them; [#152](https://github.com/grimmerk/codev/pull/152) made a row carry up to four (`match #N` plus `match path`, `match assistant`, and `match recap` or `match reply`), and at that count the labels competed with the text they point at and could not be told apart from it. They are now amber-outlined pills, the shape every other badge on the row already uses (PR, account, terminal), so solid amber means exactly one thing: text the query matched. The `id 4ed7505a` marker follows, keeping the corner radius of the badges beside it on that line. No row is taller for it: the border is drawn on an inline element, which costs no line height

README.md

@@ -34,7 +34,7 @@ Bare words search everything; **operators aim the query** (the `?` chip beside t
 | `after:7d` `after:2026-09-01` `after:today` `before:…` | by the session's last activity (`Nh` `Nd` `Nw`, a date, `today`, `yesterday`) |
 | `#137` `pr:137` `owner/repo#137` `owner/repo/pull/137` `https://github.com/owner/repo/pull/137` | **a pull request in any spelling** — the query form and the form in the text no longer have to agree (measured: 80.6% of PR mentions in prompts used only one form). Three levels of strictness; see *Finding a pull request* below |
 
-Every term must hold. An operator with an unreadable value (`after:soon`) is reported under the box and ignored rather than silently matching nothing.
+Every term must hold. A space after the colon is fine — `title: ci` reads the same as `title:ci` — unless the next token is already a search term on its own, i.e. another operator or a PR reference: `title: is:live` and `title: #137` stay two terms rather than quietly becoming a title, and the `title:` is reported. Quote it to mean it literally (`title: "is:live"`), with or without the space. An operator with an unreadable value (`after:soon`) is reported under the box and ignored rather than silently matching nothing.
 
 A result also says **why and when**: the `match #N` line steps through a session's prompt hits, up to 20 of them (`‹ 2/12 ›`) and unfolds (`▸`) the prompt before and after the hit; a `by match` chip orders results by when the match happened rather than by the session's last activity; and when the matching field is not on the row — the project path, something the assistant said, a recap the row is not showing — a `match path` / `match assistant` / `match recap` / `match reply` line names it. Fields that are on the row (title, branch, project name, PR badge, first/last prompt) already carry the highlight and add no line.
 

package.json

@@ -1,7 +1,7 @@
 {
   "name": "CodeV",
   "productName": "CodeV",
-  "version": "1.0.90",
+  "version": "1.0.91",
   "description": "Quick switcher for VS Code, Cursor, and Claude Code sessions",
   "repository": {
     "type": "git",

src/session-search.test.ts

@@ -475,6 +475,78 @@ describe('parseQuery', () => {
     expect(q.words).toEqual(['ok']);
   });
 
+  it('reads a space after the colon as part of the operator', () => {
+    const q = parseQuery('title: ci time', now);
+    expect(q.fields).toEqual([{ field: 'title', value: 'ci' }]);
+    expect(q.words).toEqual(['time']);
+    expect(q.ignored).toEqual([]);
+    // Identical to the no-space form, which is the whole point.
+    expect(q).toEqual(parseQuery('title:ci time', now));
+  });
+
+  it('takes a quoted phrase after the space, for every operator there is', () => {
+    expect(parseQuery('title: "two words" rest', now).fields).toEqual([
+      { field: 'title', value: 'two words' },
+    ]);
+    expect(parseQuery('pr: 147', now).prRefs).toEqual([
+      { number: 147, strict: true },
+    ]);
+    expect(parseQuery('has: pr', now).has).toEqual(['pr']);
+    expect(parseQuery('is: live', now).is).toEqual(['live']);
+    expect(parseQuery('after: 7d', now).after).toBe(now - 7 * 24 * 3600 * 1000);
+    expect(parseQuery('before: 3d', now).before).toBe(
+      now - 3 * 24 * 3600 * 1000,
+    );
+    // Every scoped field, so a change to SCOPED_FIELDS cannot silently leave
+    // one of them out of the rule.
+    const scoped = [
+      'title',
+      'branch',
+      'msg',
+      'project',
+      'account',
+      'recap',
+    ] as const;
+    for (const field of scoped) {
+      expect(parseQuery(`${field}: x`, now).fields).toEqual([
+        { field, value: 'x' },
+      ]);
+    }
+  });
+
+  it('never takes a token that is already a search term, and reports what was typed', () => {
+    const q = parseQuery('title: is:live', now);
+    expect(q.fields).toEqual([]);
+    expect(q.is).toEqual(['live']);
+    expect(q.ignored).toEqual(['title:']);
+    // A PR reference is a search term too: absorbing it would delete the
+    // search silently, since any non-empty value is a legal field value.
+    const pr = parseQuery('title: #137', now);
+    expect(pr.fields).toEqual([]);
+    expect(pr.prRefs).toEqual([{ number: 137 }]);
+    expect(pr.ignored).toEqual(['title:']);
+    // Nothing to take at the end of the query — someone still typing.
+    expect(parseQuery('ci title:', now).ignored).toEqual(['title:']);
+    // When the taken token is unusable, the report shows both halves.
+    expect(parseQuery('after: soon', now).ignored).toEqual(['after: soon']);
+    // An unknown key stays a bare word and takes nothing.
+    const unknown = parseQuery('error: ci', now);
+    expect(unknown.words).toEqual(['error:', 'ci']);
+    expect(unknown.ignored).toEqual([]);
+  });
+
+  it('quotes are the escape hatch, with or without the space', () => {
+    for (const literal of ['is:live', '#137', 'pr:9']) {
+      const spaced = parseQuery(`title: "${literal}"`, now);
+      expect(spaced.fields).toEqual([{ field: 'title', value: literal }]);
+      expect(spaced.is).toEqual([]);
+      expect(spaced.prRefs).toEqual([]);
+      expect(spaced.ignored).toEqual([]);
+      // The no-space form has always meant this; the two must not diverge.
+      expect(spaced).toEqual(parseQuery(`title:"${literal}"`, now));
+    }
+  });
+
   it('keeps a leading-zero hash as a bare word rather than reading it as a PR', () => {
     const q = parseQuery('#012', now);
     expect(q.prRefs).toEqual([]);

src/session-search.ts

@@ -108,31 +108,49 @@ export const isEmptyQuery = (q: ParsedQuery): boolean =>
   q.after === undefined &&
   q.before === undefined;
 
+interface QueryToken {
+  text: string;
+  /**
+   * Some part of this token was inside double quotes. The quotes are gone by
+   * the time anything reads the token, but whether they were there is the
+   * language's "treat this as literal text" mark, which the operator rule in
+   * `parseQuery` needs: `title: "is:live"` is a title, not a live filter.
+   */
+  quoted: boolean;
+}
+
 /**
  * Split on whitespace, honouring double quotes: `title:"foo bar"` is one
  * token with the quotes removed (`title:foo bar`). An unterminated quote runs
  * to the end of the query, which is what someone still typing expects.
  */
-export const tokenizeQuery = (query: string): string[] => {
-  const out: string[] = [];
+const tokenizeQueryDetailed = (query: string): QueryToken[] => {
+  const out: QueryToken[] = [];
   let cur = '';
+  let inQuotes = false;
   let quoted = false;
   for (const ch of query) {
     if (ch === '"') {
-      quoted = !quoted;
+      inQuotes = !inQuotes;
+      quoted = true;
       continue;
     }
-    if (!quoted && /\s/.test(ch)) {
-      if (cur) out.push(cur);
+    if (!inQuotes && /\s/.test(ch)) {
+      if (cur) out.push({ text: cur, quoted });
       cur = '';
+      quoted = false;
       continue;
     }
     cur += ch;
   }
-  if (cur) out.push(cur);
+  if (cur) out.push({ text: cur, quoted });
   return out;
 };
 
+/** The tokens alone, for callers that do not care how they were written. */
+export const tokenizeQuery = (query: string): string[] =>
+  tokenizeQueryDetailed(query).map((t) => t.text);
+
 // A GitHub owner is alphanumerics and hyphens only — no dot — which is what
 // keeps `example.com/o/pull/1` from reading as owner `example.com`.
 const OWNER = '[a-z0-9-]+';
@@ -223,14 +241,61 @@ export const parseQueryDate = (value: string, now: number): number | null => {
   return null;
 };
 
+/** Every `key:` the parser understands, for the space-after-colon rule below. */
+const OPERATOR_KEYS: ReadonlySet<string> = new Set([
+  ...SCOPED_FIELDS,
+  'pr',
+  'has',
+  'is',
+  'after',
+  'before',
+]);
+
+/** Is this token itself an operator (`is:live`), rather than a plain value? */
+const isOperatorToken = (token: string): boolean => {
+  const colon = token.indexOf(':');
+  return colon > 0 && OPERATOR_KEYS.has(token.slice(0, colon));
+};
+
+/**
+ * May this token become the value of the operator before it?
+ *
+ * A QUOTED token always may: quotes are the language's mark for literal text,
+ * so `title: "is:live"` is a title of `is:live`, matching what the no-space
+ * `title:"is:live"` has always done.
+ *
+ * An unquoted one may not when it is already a search term in its own right —
+ * another operator, or a PR reference. Those meant something before this rule
+ * existed, and absorbing them would silently delete a term the user asked for:
+ * `title: #137` used to search for PR 137 (and report `title:` as unusable),
+ * so it still does, rather than quietly becoming a title of `#137`.
+ */
+const isTakeableValue = (token: QueryToken): boolean => {
+  if (token.quoted) return true;
+  const lower = token.text.toLowerCase();
+  return !isOperatorToken(lower) && !parsePrRef(lower);
+};
+
 /**
  * Parse the search box. Everything is lowercased; a token with an unknown
  * `key:` prefix (`error:`, `12:30`, a URL that is not a PR) stays a bare
  * word, so the operators cost nothing to queries that do not use them.
+ *
+ * A space after the colon is allowed: `title: ci` reads as `title:ci`. People
+ * type the space (it is how a sentence works, and how most search boxes read),
+ * and a bare `title:` meant nothing before — it was reported as an unreadable
+ * value — so for a plain word the rule only changes queries that were already
+ * errors. Which token may be taken is `isTakeableValue`: quoted text always,
+ * and anything that is not itself a search term (an operator, a PR reference)
+ * otherwise. So `title: is:live` and `title: #137` keep doing what they did,
+ * each with its `title:` reported, and `title: "is:live"` is the way to ask
+ * for that literal title — the same escape hatch as `title:"is:live"`.
  */
 export const parseQuery = (query: string, now = Date.now()): ParsedQuery => {
   const q = emptyQuery();
-  for (const raw of tokenizeQuery(query)) {
+  const tokens = tokenizeQueryDetailed(query);
+  for (let i = 0; i < tokens.length; i++) {
+    const raw = tokens[i].text;
     const token = raw.toLowerCase();
     const pr = parsePrRef(token);
     if (pr) {
@@ -239,28 +304,40 @@ export const parseQuery = (query: string, now = Date.now()): ParsedQuery => {
     }
     const colon = token.indexOf(':');
     const key = colon > 0 ? token.slice(0, colon) : '';
-    const value = colon > 0 ? token.slice(colon + 1) : '';
+    let value = colon > 0 ? token.slice(colon + 1) : '';
+    // What the warning line shows when the value turns out to be unusable:
+    // both tokens when the next one was taken, so `after: soon` is reported
+    // as the user typed it rather than as a bare `after:`.
+    let shown = raw;
+    if (key && !value && OPERATOR_KEYS.has(key)) {
+      const next = tokens[i + 1];
+      if (next !== undefined && isTakeableValue(next)) {
+        value = next.text.toLowerCase();
+        shown = `${raw} ${next.text}`;
+        i++;
+      }
+    }
     if (key === 'pr') {
       // `pr:147`, `pr:o/r#147`, `pr:<url>` — a number alone is allowed here
       // because the key already says what it is.
       const ref = /^[1-9][0-9]*$/.test(value)
         ? { number: Number(value), strict: true }
         : parsePrRef(value);
       if (ref) q.prRefs.push(ref);
-      else q.ignored.push(raw);
+      else q.ignored.push(shown);
     } else if (SCOPED_FIELDS.has(key)) {
       if (value) q.fields.push({ field: key as ScopedField, value });
-      else q.ignored.push(raw);
+      else q.ignored.push(shown);
     } else if (key === 'has') {
       if (HAS_VALUES.has(value)) q.has.push(value);
-      else q.ignored.push(raw);
+      else q.ignored.push(shown);
     } else if (key === 'is') {
       const v = IS_ALIASES[value] ?? value;
       if (IS_VALUES.has(v)) q.is.push(v);
-      else q.ignored.push(raw);
+      else q.ignored.push(shown);
     } else if (key === 'after' || key === 'before') {
       const t = parseQueryDate(value, now);
-      if (t === null) q.ignored.push(raw);
+      if (t === null) q.ignored.push(shown);
       else if (key === 'after') q.after = Math.max(q.after ?? -Infinity, t);
       else q.before = Math.min(q.before ?? Infinity, t);
     } else if (token) {