) * Add a test suite, migrate to vite/vitest, and fix the match counts The repo had no working tests: jest and ts-jest were installed with no config and no test script, and the single test file was an empty stub. Nothing typechecked or linted on CI either, and the Cypress job had been failing on every run because wordpress.cy.ts needs a dockerised WordPress that CI never starts. Tooling - Replace jest with vitest (jsdom) and webpack with vite; one IIFE bundle per entry, as an MV3 content script cannot use ESM imports - Add test/lint/format/typecheck/checks scripts and a CI checks job - `checks` ends with a build, so dist cannot silently go stale - Exclude wordpress.cy.ts from the default e2e run; it has its own script - TypeScript 4.4 -> 5.9; three tsconfigs now typecheck, including the Cypress specs, whose config listed "dom" as an @types package and so failed outright once anything checked it Tests - 614 unit tests, up from zero, covering regex, util, elements, hints, history, the background listeners and searchReplace itself - searchreplace.invariants.test.ts asserts, over a matrix of page shapes and option sets, that `replaced` never exceeds `original` - counts.cy.ts covers the same ground in a real browser, where CSS actually applies Count fixes (the "-2 matches" report) - ELEMENT_FILTER was unanchored, so HEADER matched HEAD and every site's <header> was dropped from the count while still being replaced - In HTML mode the count came from a filtered clone while the replacement rewrote the original element's innerHTML, so more was replaced than was counted; the count now comes from the same string that is rewritten - getRemainingMatchCount clamps the displayed figure at zero Other fixes - listenerContentResponse fired checkPreviousResponse without awaiting, so the background replied before merging iframe results - install.ts asked the background script for the language list from the background script, which MV3 never delivers, so detection always fell back to English - saved.ts dereferenced result.storage.saved unguarded - waitForIframeLoad had no timeout and could hang the whole search - History was capped at 10 before deduplication, dropping distinct entries -   is now matched in text mode, where the DOM holds U+00A0 (#1) - The content script is injected into open tabs on install and update, so no reload is needed (#102); needs the new `scripting` permission - Enter runs the replacement and Shift+Enter inserts a newline, so multi-line terms can be typed (#28) - Drop popup.js, options.js, help.js, util.js and elements.js from content_scripts; only searchreplace.js is a content script - noImplicitAny enabled and the resulting 26 errors fixed; searchReplace takes a typed options object rather than 14 positional arguments UI - Revert the green-and-white colours to the original greys, keeping the redesign's layout; colours are now custom properties - Restore the original grey magnifying-glass icons - Consistent spacing scale, aligned panel gutters, no fixed-height body panel, buttons that no longer wrap, and no empty footer band Known limitations, pinned by tests named KNOWN LIMITATION: Replace HTML ignores the "Hidden content" checkbox and rewrites a script that shares a container with matching text, and its pre-replace count under-reports. All three need the excluded subtrees preserved across the rewrite. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * Fix CI: Node 24 for jsdom, and least-privilege workflow permissions The checks job failed on Node 20: jsdom 30 reaches for webidl.util.markAsUncloneable in undici, which that Node does not have. Both workflows now use Node 24, matching what the test suite is developed against, and package.json records the floor in `engines`. CodeQL also flagged the new workflow for not restricting the default GITHUB_TOKEN, so both now declare `contents: read`. Its two js/xss-through-dom findings are the pre-existing Replace HTML data flow, which was previously invisible to analysis behind a computed property assignment. Documented at setReplaceTarget rather than changed, as re-parsing the page's own markup is what the option does. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * Note the Node version requirement in the README Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Cannot replace &nbsp; in HTML #1

Description

Activity

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

Cannot replace in HTML #1