fix(issue-ref): do not read a URL fragment as owner/repo#12

changelog.d/2435-issue-ref-url-fragment.fixed.md

@@ -0,0 +1 @@
+Reading an issue number out of a line no longer turns a link's URL fragment into an issue reference: `https://example.com/page#123` used to name issue #123 and write a card for an issue nobody meant, while `owner/repo#12`, a bare `#12` and GitHub issue and pull-request links keep naming the issue they point at ([#2435](https://github.com/ThinkFlowLab/FlyBot/issues/2435)).

core/src/shared/issue-ref.js

@@ -6,8 +6,10 @@
 // front makes it part of an identifier, a version, a hex colour or a URL fragment (`C#7`, `1.2#3`,
 // `#3366ff`, `help/#123`), and a word character behind means the number keeps going (`#1360abc`).
 // `owner/repo#12` is its own shape, tried separately, so this separation cannot swallow it: there
-// the `#` follows a repo name rather than a space.
-const ISSUE_REF = /(?<![\w./-])#(\d{1,7})(?![\w])|(?:github\.com\/)?([\w.-]+)\/([\w.-]+)#(\d{1,7})(?![\w])|github\.com\/([\w.-]+)\/([\w.-]+)\/(?:issues|pull)\/(\d{1,7})(?![\w])/gi
+// the `#` follows a repo name rather than a space. It needs the same boundary in front, or a link
+// whose path carries a segment before the `#` reads as that shape and names an issue nobody meant
+// (`https://example.com/page#123`, #1392).
+const ISSUE_REF = /(?<![\w./-])#(\d{1,7})(?![\w])|(?<![\w./-])(?:github\.com\/)?([\w.-]+)\/([\w.-]+)#(\d{1,7})(?![\w])|github\.com\/([\w.-]+)\/([\w.-]+)\/(?:issues|pull)\/(\d{1,7})(?![\w])/gi
 
 // The id a match of `ISSUE_REF` names, normalised to `#1234`. The three shapes carry their number in
 // different groups (1, 4, 7); every reader takes it from here, so no caller can pick the wrong one.

core/test/issue-ref-url-fragment.test.js

@@ -0,0 +1,42 @@
+// A URL fragment is not an issue reference (#1392). The repository-qualified shape `owner/repo#12`
+// had no boundary in front of it, so a link whose path carries a segment before the `#` read as
+// `<host>/<segment>#12`: `https://example.com/page#123` named issue #123 and the dispatch wrote a
+// card for an issue nobody meant (#1360). The boundary and the number are asserted apart: a
+// fragment must not hand over digits, and a real GitHub URL with a fragment must still yield exactly
+// one number.
+import assert from 'node:assert/strict'
+import { test } from 'node:test'
+import { issueRefs, namesIssue } from '../src/shared/issue-ref.js'
+
+const fragments = [
+  ['a path segment before the #', 'see https://example.com/page#123 for details'],
+  ['a two-segment path', 'https://example.com/docs/guide#12'],
+  ['a nested path', 'read https://docs.example.org/a/b#42 then'],
+  ['a fragment of seven digits', 'see https://example.com/page#1234567 for details'],
+  ['a fragment longer than an id', 'https://example.com/page#12345678'],
+]
+
+for (const [what, text] of fragments) {
+  test(`a URL fragment names no issue (${what})`, () => {
+    assert.deepEqual(issueRefs(text), [], text)
+  })
+}
+
+test('a fragment does not keep the digits of the link it follows', () => {
+  assert.equal(namesIssue('https://example.com/page#123', 123), false)
+  assert.equal(namesIssue('https://example.com/page#1234', 123), false)
+})
+
+test('one GitHub URL yields exactly one issue number, fragment and all', () => {
+  assert.deepEqual(issueRefs('https://github.com/o/r/issues/123#issuecomment-4567'), ['#123'])
+  assert.deepEqual(issueRefs('https://github.com/o/r/pull/9#discussion_r123456'), ['#9'])
+})
+
+test('the shapes that do name an issue keep working beside a URL', () => {
+  assert.deepEqual(issueRefs('owner/repo#1360'), ['#1360'])
+  assert.deepEqual(issueRefs('see owner/repo#1360 for the fix'), ['#1360'])
+  assert.deepEqual(issueRefs('please rebase #1360 onto main'), ['#1360'])
+  assert.deepEqual(issueRefs('https://github.com/o/r/issues/1360'), ['#1360'])
+  assert.deepEqual(issueRefs('@Dev-2 rebase owner/repo#1360'), ['#1360'])
+  assert.equal(namesIssue('owner/repo#1360', 1360), true)
+})

patches/ds-bot/files/src/shared/issue-ref.js

@@ -6,8 +6,10 @@
 // front makes it part of an identifier, a version, a hex colour or a URL fragment (`C#7`, `1.2#3`,
 // `#3366ff`, `help/#123`), and a word character behind means the number keeps going (`#1360abc`).
 // `owner/repo#12` is its own shape, tried separately, so this separation cannot swallow it: there
-// the `#` follows a repo name rather than a space.
-const ISSUE_REF = /(?<![\w./-])#(\d{1,7})(?![\w])|(?:github\.com\/)?([\w.-]+)\/([\w.-]+)#(\d{1,7})(?![\w])|github\.com\/([\w.-]+)\/([\w.-]+)\/(?:issues|pull)\/(\d{1,7})(?![\w])/gi
+// the `#` follows a repo name rather than a space. It needs the same boundary in front, or a link
+// whose path carries a segment before the `#` reads as that shape and names an issue nobody meant
+// (`https://example.com/page#123`, #1392).
+const ISSUE_REF = /(?<![\w./-])#(\d{1,7})(?![\w])|(?<![\w./-])(?:github\.com\/)?([\w.-]+)\/([\w.-]+)#(\d{1,7})(?![\w])|github\.com\/([\w.-]+)\/([\w.-]+)\/(?:issues|pull)\/(\d{1,7})(?![\w])/gi
 
 // The id a match of `ISSUE_REF` names, normalised to `#1234`. The three shapes carry their number in
 // different groups (1, 4, 7); every reader takes it from here, so no caller can pick the wrong one.

patches/ds-bot/files/test/issue-ref-url-fragment.test.js

@@ -0,0 +1,42 @@
+// A URL fragment is not an issue reference (#1392). The repository-qualified shape `owner/repo#12`
+// had no boundary in front of it, so a link whose path carries a segment before the `#` read as
+// `<host>/<segment>#12`: `https://example.com/page#123` named issue #123 and the dispatch wrote a
+// card for an issue nobody meant (#1360). The boundary and the number are asserted apart: a
+// fragment must not hand over digits, and a real GitHub URL with a fragment must still yield exactly
+// one number.
+import assert from 'node:assert/strict'
+import { test } from 'node:test'
+import { issueRefs, namesIssue } from '../src/shared/issue-ref.js'
+
+const fragments = [
+  ['a path segment before the #', 'see https://example.com/page#123 for details'],
+  ['a two-segment path', 'https://example.com/docs/guide#12'],
+  ['a nested path', 'read https://docs.example.org/a/b#42 then'],
+  ['a fragment of seven digits', 'see https://example.com/page#1234567 for details'],
+  ['a fragment longer than an id', 'https://example.com/page#12345678'],
+]
+
+for (const [what, text] of fragments) {
+  test(`a URL fragment names no issue (${what})`, () => {
+    assert.deepEqual(issueRefs(text), [], text)
+  })
+}
+
+test('a fragment does not keep the digits of the link it follows', () => {
+  assert.equal(namesIssue('https://example.com/page#123', 123), false)
+  assert.equal(namesIssue('https://example.com/page#1234', 123), false)
+})
+
+test('one GitHub URL yields exactly one issue number, fragment and all', () => {
+  assert.deepEqual(issueRefs('https://github.com/o/r/issues/123#issuecomment-4567'), ['#123'])
+  assert.deepEqual(issueRefs('https://github.com/o/r/pull/9#discussion_r123456'), ['#9'])
+})
+
+test('the shapes that do name an issue keep working beside a URL', () => {
+  assert.deepEqual(issueRefs('owner/repo#1360'), ['#1360'])
+  assert.deepEqual(issueRefs('see owner/repo#1360 for the fix'), ['#1360'])
+  assert.deepEqual(issueRefs('please rebase #1360 onto main'), ['#1360'])
+  assert.deepEqual(issueRefs('https://github.com/o/r/issues/1360'), ['#1360'])
+  assert.deepEqual(issueRefs('@Dev-2 rebase owner/repo#1360'), ['#1360'])
+  assert.equal(namesIssue('owner/repo#1360', 1360), true)
+})