## Problem Rspack is already the default bundler for new GDK setups: `rspack.enabled` in `components/rspack/GDK.rb` computes to `!vite.enabled? && !webpack.enabled?`, so anyone with no bundler keys in `gdk.yml` is on Rspack today. The problem is what happens to everyone else. `GDK::RspackMigration` copies `webpack.*` tuning keys onto `rspack.*` and deletes `webpack.enabled` and `vite.enabled` from `gdk.yml`, and until recently it ran on every `gdk update` with no date gate. A user who deliberately set `webpack.enabled true`, the documented fallback in `data/announcements/0048_rspack_default_bundler.yml`, was switched back to Rspack on their next update, and frontend engineers who turned on Vite to evaluate its bundling progress lost the setting the same way. 4a58ee45e stopped that by skipping the task outright, which leaves the opposite problem: `gdk rake gdk:migrate:rspack` no longer migrates anyone who asks for it, because `skip!` raises before the migration runs. There is also no clean way to switch bundlers, or to find out you are on a retired one. Right now people hand-edit `gdk.yml` through `gdk config set` and have to remember `gdk reconfigure` and a service restart themselves. `GDK::Command.check_unsupported_bundler` warns about Vite on every invocation and says nothing about Webpack. This plan stops the silent revert, and removes Webpack only after we have announced the date, given people time to react, and checked the numbers. ## Plan Auto-migration stays off until a fixed cutoff. Users get told early and repeatedly. Switching is easy in both directions. Cutoff: **19 October 2026**. This needs no new commands. The existing `gdk rake gdk:migrate:rspack` task migrates forward, and `gdk config set vite.enabled true`, or `gdk config set webpack.enabled true` for anyone who never used Vite, moves back. The only reason moving back does not work today is the ungated automatic migration. ## Phase 1 - before the cutoff (by 22 September 2026) Date-gating the migration is what makes switching back work. Once `gdk:migrate:rspack` stops running on every `gdk update`, `gdk config set webpack.enabled true` becomes a setting that sticks, which is all a rollback ever needed to be. No new machinery. Switching to Rspack on demand is `gdk rake gdk:migrate:rspack`, which already exists and already moves `webpack.*` tuning keys onto `rspack.*`. GDK's convention is to expose maintenance migrations this way - the asdf error message in `lib/gdk/command.rb` points people at `gdk rake mise:migrate` in exactly the same way. It needs documenting and printing in the warning, not rewriting. As with any config change, `gdk reconfigure` and `gdk restart` follow. Switching back leads with Vite, since Vite survives the cutoff and Webpack does not: `gdk config set vite.enabled true`, then `gdk reconfigure` and `gdk restart`. Webpack is the last resort, for anyone Vite does not work for or who never used it, and only until the cutoff. Someone already on Vite is never sent to Webpack. `GDK::Command::Config` already prints the reconfigure reminder after every `set`, so this needs no special case. - [x] Add a cutoff constant to `GDK::RspackMigration`, drop the blanket `t.skip!`, and leave `migrate:rspack` out of the `gdk:migrate` chain until that date, so `gdk update` stops reverting deliberate Webpack and Vite choices while the task still works on demand. Merge request !6143. - [x] After `GDK::RspackMigration` migrates, print the exact commands to switch back, so the way out is on screen at the moment someone needs it. `remove_superseded_keys` already returns the keys it deleted: a Vite user is offered Vite alone, and a Webpack user is offered Vite first with Webpack as the last resort until the cutoff. Merge request !6143. - [x] Replace the Vite-only banner in `check_unsupported_bundler` with a warning that covers Webpack and Vite. It must state the cutoff date, the forward command (`gdk rake gdk:migrate:rspack`), the way back (`gdk config set vite.enabled true` first, `gdk config set webpack.enabled true` as the last resort, then `gdk reconfigure` and `gdk restart`), and a link to this issue for reports. Drive the cadence with `GDK::ReminderHelper`, the same helper `gdk:migrate:mise` uses for its 5-day reminder, instead of warning on every invocation, and shorten the interval as 19 October approaches. - [x] Send a telemetry event when a non-Rspack bundler is detected, so we can see how many people are still on Webpack or Vite. Precedent: the `asdf-error` custom event in `lib/gdk/command.rb`. - [x] Add announcement `data/announcements/0049_*.yml` stating the cutoff date and both commands. `0048` and `0045` are each already used twice, so confirm the next free number first. Announcements render once, on install or update, so this supplements the reminders rather than replacing them. - [x] Document the switch, the way back with Vite first and Webpack as the last resort, and the cutoff in `doc/configuration.md` (Rspack, Vite and Webpack sections) and `doc/troubleshooting/rspack.md`. ## Phase 2 - notice window (22 September to 19 October 2026) - [ ] Ask frontend engineers to run `gdk rake gdk:migrate:rspack` early, while backing out still works, so bugs surface before the cutoff instead of after it. - [ ] Triage reports in this issue and fix the Rspack bugs they surface. - [ ] Check the telemetry before the cutoff. If a meaningful number of people are still on Webpack, or Rspack bugs are still unresolved, move the date instead of shipping the removal on schedule. ## Phase 3 - after the cutoff (from 19 October 2026) - [ ] Flip the date gate so `gdk:migrate:rspack` migrates automatically again on `gdk update`. - [ ] Change `SUPERSEDED_KEYS` in `lib/gdk/rspack_migration.rb` so the automatic migration stops deleting `vite.enabled`. Vite stays available for frontend engineers evaluating its bundling progress, behind the unsupported-bundler warning. Force-reverting it on every update would break that use case. - [ ] Remove Webpack as a GDK option. Touchpoints: - `components/webpack/GDK.rb` - the `!config.webpack.enabled?` term in `components/rspack/GDK.rb` - `lib/gdk/services/webpack.rb` - the `'webpack' => 'KILL'` entry in `lib/runit/config.rb` - the two `config.webpack.host` / `config.webpack.port` fallbacks in `support/templates/gitlab/config/gitlab.yml.erb` - the Vite/Webpack conflict check in `lib/gdk/services/vite.rb` - `lib/gdk/diagnostic/frontend_dev_server.rb` - `lib/gdk/rspack_migration.rb` - regenerate `gdk.example.yml` with `rake gdk.example.yml`, since `make verify-gdk-example-yml` enforces it - docs: `doc/configuration.md`, `doc/troubleshooting/webpack.md` and its `_index.md` entry, `doc/advanced.md`, `doc/troubleshooting/running_tests.md` - the Webpack expectations in `spec/lib/gdk/config_spec.rb` - [ ] Drop Webpack from the unsupported-bundler warning and from the docs, so the remaining warning only covers Vite. From then on the only switch left is `gdk config set vite.enabled true`, for testing Vite's bundling progress. - [ ] Decide whether the migration should delete the whole `webpack` block instead of only `webpack.enabled`. Leftover `webpack:` keys are harmless once the component is gone, since `ConfigSettings#validate!` only walks defined attributes, but a clean removal is worth doing. ## Scope This retires Webpack as a GDK service. It does not remove Webpack from `gitlab-org/gitlab`, which still defaults to Webpack in `yarn build` behind the `ENABLE_RSPACK` flag and still runs a `webpack-dev-server` CI job. Anyone who needs Webpack after the cutoff can still run `yarn dev-server` in the `gitlab` directory by hand. Link the `gitlab`-side removal effort here so both timelines stay in sync. ## Reporting problems Report problems with `gdk rake gdk:migrate:rspack`, or with Rspack itself, in this issue. The warning links here directly. ## Known issues (as of 2026-10-01) Problems found during the notice window. Keep this list up to date. Please report new ones in this issue. | Problem | Tracking | Fix | Status | | --- | --- | --- | --- | | Rspack crash-loops on a broken persistent cache (`Dependency with ID ... not found`) | https://gitlab.com/gitlab-org/gitlab-development-kit/-/work_items/3373 | https://gitlab.com/gitlab-org/gitlab/-/merge_requests/258952 (update Rspack to 2.2.8). Verified with the upstream repro and on a GDK dev server with an emptied cache `meta/`. Upstream: https://github.com/web-infra-dev/rspack/issues/14863 | MR open. 3 reports. More urgent since https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6176 (stop with KILL) merged first. | | First use of a lazily compiled chunk stalls the UI, `XML Parsing Error` from `_rspack/lazy/trigger` | https://gitlab.com/gitlab-org/gitlab-development-kit/-/work_items/3374 | None yet | Needs triage. | | Orphaned `runsv` keeps `supervise/lock`, so the new `runsvdir` never supervises rspack | Reported on the epic, no issue: https://gitlab.com/groups/gitlab-org/-/work_items/22569#note_3771687732 | https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6126 (wait for runsv on stop, merged 2026-09-16, prevents new orphans but does not clean up existing ones) and https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6127 (`gdk doctor --correct` for stale services, open) | Partly fixed. | | Bundler child exits but the service stays "up" and nothing listens on port 3808 | Epic report, no issue | https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6128 (report-only `gdk doctor` check, open since 2026-09-10) | In review. | | Crash loop on an Apple Silicon Mac (`EMFILE` watcher errors, then silent SIGBUS) | https://gitlab.com/gitlab-org/gitlab-development-kit/-/work_items/3361 | None yet | Needs triage, no response since 2026-09-15. | | Serving GDK on a LAN IP address ([doc](https://gitlab.com/gitlab-org/gitlab-development-kit/-/blob/main/doc/howto/local_network.md#access-gdk-on-private-ip-address)) with Rspack. The same reporter's Vite setup also broke recently. | https://gitlab.com/gitlab-org/gitlab-development-kit/-/work_items/3349#note_3919800597 | None. `rspack.host` defaults to `listen_address`, so it is expected to work, but nobody has confirmed it | Waiting for details from the reporter. | ### Unconfirmed reports - Several people in Slack say Rspack "still doesn't work" and stayed on Vite, without details: https://gitlab.slack.com/archives/C02PF508L/p1790333885980599 , https://gitlab.slack.com/archives/C02PF508L/p1790573627693339 . Ask them to report here with `gdk doctor` output and the rspack log. - A stale `gitlab/public/assets/webpack` from an earlier static build caused a manifest error after switching. Deleting it fixed it. One report: https://gitlab.slack.com/archives/C02PF508L/p1790333885980599 ### Fixed - `rspack serve` orphaned on stop/restart, EADDRINUSE restart loop: https://gitlab.com/gitlab-org/gitlab-development-kit/-/work_items/3371, https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6176, https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6177 - HMR port with `gitlab_http_router`: https://gitlab.com/gitlab-org/gitlab-development-kit/-/work_items/3365, https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6157 - Open file limit: https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6125 - Bind address: https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6137 - Bundler in `gitlab.yml`: https://gitlab.com/gitlab-org/gitlab-development-kit/-/merge_requests/6130