fedi-patchset

Build the fedi.my.id Mastodon fork on top of a current glitch-soc checkout, from a stack of ten patches.

• Base commit: .base-commit — glitch-soc main 29468cc4c6427395eb99218706542b1a89e43720 (2026-11-03, v4.7.5)
• Result: 10 patches, 486 files changed, +55291 / −318 against that commit
• Licence: AGPL-3.0, like glitch-soc and Mastodon — see LICENSE

Why this exists

The fork is glitch-soc plus three upstream feature branches by TheEssem (emoji reactions, bubble timeline, GIF picker) plus the operator's own customisations:

glitch-soc/mastodon ─┬─ TheEssem/feature/reaction-list ───┐
                     ├─ TheEssem/feature/bubble-timeline ─┼─ fedi.my.id @dev
                     └─ TheEssem/feature/gif-picker-v2 ───┘        │
                                                                   └─ operator customisations

Historically those were assembled by merging neatchee/mastodon, which meant new Mastodon/glitch-soc releases could only be picked up after that re-merge. This repository keeps the same result as a patch series that applies directly to glitch-soc main, so upstream updates are a rebase here instead of a wait.

Since 2026-10-06 the fork's dev branch is this result: commit 43e2b637c8, whose second parent is the series (glitch-soc b3877d5b24 plus patches 1–5) and whose tree is the verified build tree b473d8b76fe2fb8cd5084e3135a62ea923c2d126. It was pushed as a fast-forward over the previous dev (b416d280e0), which is preserved as the branch backup-20261006. Wherever these documents say "dev", they mean that pre-sync branch unless they say otherwise; docs/fork-comparison.md explains what the comparison now checks. The fork's default branch main still points at the pre-sync merge history (0828922071), so a fresh clone gets the fork before this rebase — check out dev for the current tree.

Requirements

For	Needs
building the tree, comparing it with the fork	bash ≥ 4, git, coreutils
scripts/setup.sh --stg	Stacked Git 2.x (stg)
the app's own checks (format:check, typecheck, lint, test:js)	Node ≥ 22 and Yarn 4
the container image (scripts/build-image.sh, the image workflow)	Docker with buildx
Ruby-side checks (bin/rubocop, haml-lint, i18n-tasks, migrations, specs)	Ruby 4.0.7 + the app's gems, PostgreSQL, Redis

Quick start

# 1. build a working tree: clone glitch-soc, check out the base commit, apply the series
scripts/setup.sh --dest ~/src/fedi.my.id          # plain commits (git am)
scripts/setup.sh --dest ~/src/fedi.my.id --stg    # or a Stacked Git stack

# 2. install and configure the app
cd ~/src/fedi.my.id
bundle install && yarn install
cp .env.production.sample .env.production         # then edit it
RAILS_ENV=production bundle exec rails db:setup

# 3. check the build
scripts/compare-with-fork.sh --result ~/src/fedi.my.id   # vs fedi.my.id@dev
BASE_REPO=~/src/mastodon scripts/lint-patches.sh         # patches are internally sound

Since the 2026-10-06 sync dev holds exactly the build tree, so the comparison above is a mirror check; add --fork-ref backup-20261006 to compare against the fork as it was before the sync — that is what the numbers in docs/fork-comparison.md describe.

setup.sh options: --dest DIR, --branch NAME, --remote URL, --stg, --force, --help. It is safe to re-run: an existing checkout is reused, the branch is reset to the base commit and the series is applied again.

Repository layout

.base-commit                     glitch-soc commit the series applies to
series                           patch order, read by the scripts
patches/0001..0005-*.patch       the patch series (git format-patch mail files)
LICENSE                          AGPL-3.0 (glitch-soc / Mastodon)
README.md                        this file
docs/runbook.md                  the steps the operator does by hand (rebase, dev, image, deploy, rollback)
docs/updating.md                 rebasing onto a newer glitch-soc
docs/patch-reference.md          what each patch contains and why
docs/verification.md             how to verify a build, and what was verified
docs/fork-comparison.md          how the result relates to the fork (and its dev branch)
docs/image.md                    the container image: how it is built, tagged and run
scripts/setup.sh                 build the patched tree
scripts/lint-patches.sh          static checks on the series + apply test
scripts/compare-with-fork.sh     compare the result with the fork (dev; --fork-ref)
scripts/build-image.sh           build the container image from a patched tree
scripts/check-ruby-syntax.mjs    parse changed Ruby files without Ruby (Prism/WASM)
.github/workflows/image.yml      builds and pushes ghcr.io/pegelinuxtop/fedi.my.id

The patch series

#	Patch	Files	Contents
1	feature/reaction-list	97	Emoji reactions (TheEssem): status-reaction API, service layer, federation, notifications, reaction list UI, notification/admin settings, 2 migrations + 2 post-migrations
2	feature/bubble-timeline	60	Bubble timeline (TheEssem): bubble-domain API + admin UI, BUBBLE feed and column, streaming channels, fan-out, settings, 3 migrations — plus the fork's per-timeline boost/reply settings wired through fan-out and their specs
3	feature/gif-picker	31	Tenor/Klipy GIF search (TheEssem): GIF API client + picker UI, gif_search in the instance serializers, locales
4	fedi/branding-themes	273	The operator's own customisations: sign-in banner, local-settings page, qrtool decoder, custom modern/gekka/sakura/tangerine UI themes and skins, reject-pattern settings, locale additions, 3 migrations — plus the runtime, lint and spec fixes of deviations 8–10
5	fedi/compat-overlay	14	Reconciliation with glitch-soc and with the repo's own check suite: Elk footer link, Docker qrtool stage, ja.yml blurhash strings, en.json reaction notification, eslint.config.mjs, lint fixes for five files the fork does not otherwise touch, and three upstream specs adjusted for the fork's behaviour and its root-level domain_blocks.csv

Per-patch sizes are 2391/+92−, 1476/+83−, 886/+21−, 49748/+113− and 74/+10− (lines added/removed). The split is by feature or customisation, so an upstream rebase conflicts on the furthest patch that touches a file. Shared configuration that serves more than one feature (config/settings.yml, db/schema.rb, app/models/form/admin_settings.rb, config/locales/en.yml, config/routes/admin.rb, db/migrate/*) is carried by a single patch rather than repeated. Five files legitimately receive incremental hunks from two patches (app/javascript/flavours/glitch/locales/en.json, app/serializers/rest/instance_serializer.rb, config/locales-glitch/en.yml, config/routes/api.rb, .env.production.sample: patch 1 then patch 3), and patch 1 alone references Setting.bubble_live_feed_access/bubble_topic_feed_access, which patch 2 defines — so the patches are not meant to be cherry-picked individually. See docs/patch-reference.md.

Migrations

The ten migrations the fork adds — eight under db/migrate and two under db/post_migrate — are kept verbatim, with their original filenames, timestamps and class names, and no existing migration is modified. None of their timestamps collides with a migration present at the base commit, which scripts/lint-patches.sh enforces for both directories.

0001  20221124114030_create_status_reactions.rb
0001  20240411044156_add_reaction_count_to_status_stat.rb
0001  post_migrate/20250305023754_add_reaction_counts_to_status_stat.rb
0001  post_migrate/20260520184138_normalize_status_reaction_variation_selectors.rb
0002  20240114042123_create_bubble_domains.rb
0002  20251018223804_add_bubble_timeline_preview_setting.rb
0002  20251024193240_add_bubble_timeline_topic_preview_setting.rb
0004  20221218015350_fix_foreign_keys_status_reactions.rb
0004  20230215074425_move_emoji_reaction_settings.rb
0004  20250518031405_remove_quote_id_from_statuses.rb

The container image

ghcr.io/pegelinuxtop/fedi.my.id is glitch-soc with this series applied, built by scripts/build-image.sh from the tree's own Dockerfile (fork stages included). The image workflow builds it on demand, weekly on glitch-soc main, and whenever the patches change, tagging latest, glitch-<sha12> and patchset-<rev7>. The same script builds it locally:

scripts/build-image.sh                       # pinned .base-commit, loaded locally
scripts/build-image.sh --glitch-ref main --push   # current glitch-soc, pushed

docs/image.md covers the triggers and the tag scheme, the first-time GHCR setup, multi-platform (arm64) builds, and how to run the image.

Keeping up with glitch-soc

git fetch origin main                       # in the built checkout
cd /path/to/fedi-patchset
# StGit stack (recommended — patches stay annotatable and refreshable):
(cd ~/src/fedi.my.id && stg rebase <new glitch-soc commit>)
# or plain commits:
(cd ~/src/fedi.my.id && git rebase --onto <new commit> <old base> fedi-patchset)

Then re-export the patches, update .base-commit and re-run the checks — docs/updating.md has the full procedure, including where conflicts are likely to land and the checklist to run before pushing.

Verifying a build

Script	Answers
scripts/lint-patches.sh	is the series internally sound and does it apply to .base-commit? (BASE_REPO=<glitch-soc clone> also applies it in a throwaway worktree)
scripts/compare-with-fork.sh	does the built tree still contain every fork change, and is every remaining difference explained?
scripts/check-ruby-syntax.mjs	do the changed .rb/.rake files parse? (Prism/WASM, self-testing)
the app's own scripts	yarn format:check, yarn typecheck, yarn lint:css, yarn lint:js, yarn test:js
the Ruby-side checks	bin/rubocop, bin/haml-lint, bin/i18n-tasks …, bin/rails db:migrate plus a db:schema:dump comparison, bin/flatware rspec — needs Ruby 4.0.7, PostgreSQL and Redis, see docs/verification.md

Exact commands, expected output and the one check that could not be run here (a browser session against a real server) are in docs/verification.md; the fork comparison is explained in docs/fork-comparison.md.

Deviations from the fork

The tree is generated from a merge of the fork's pre-sync dev (fedi.my.id@dev at b416d280e0, now the branch backup-20261006) onto the base commit, so it is the fork's tree plus glitch-soc's evolution since the fork's last glitch-soc merge. dev itself now holds exactly this tree, so these are the intentional differences from that pre-sync branch:

1. The fork's per-timeline boost/reply settings are now wired through fan-out (config/settings.yml, app/services/fan_out_on_write_service.rb, spec/services/fan_out_on_write_service_spec.rb). In dev the four admin toggles (show_reblogs/replies_in_{local,federated}_timelines) only affected the REST timeline: broadcastable? still required the pre-rename show_reblogs_in_public_timelines (default false in dev's config/settings.yml and exposed nowhere, so a boost reached the public streams only if a settings row happened to hold true), the reply toggle was never consulted (the corresponding expression in broadcast_to_public_streams! was a no-op), and timeline:public:bubble was published twice per bubble status. Now config/settings.yml carries the four split keys (default false), broadcast_to_public_streams! filters every channel with its own pair — timeline:public:local with _local_, and timeline:public, timeline:public:remote, timeline:public:bubble plus their :media variants with _federated_ — and self-replies keep publishing. What to expect: boosts reach the public streams again when a toggle is on, non-self replies stop streaming when the reply toggle is off, and the mixed timeline:public follows the federated pair (so a local non-self reply can appear in the mixed timeline but not in the local one). Boosts still never reach hashtag streams or hashtag followers, because a local boost is stored as a tagless wrapper (ReblogService creates it with text: '' and no tags), so @status.tags is empty. The new spec contexts pass (bin/flatware rspec spec/services/fan_out_on_write_service_spec.rb); the first version of that block was missing a let(:visibility), so it never ran — that is fixed.
2. app/javascript/flavours/glitch/initial_state.ts — dev's useSystemEmojiFont = getMeta('system_emoji_font') export was dropped: the meta field no longer exists in glitch-soc, the export is unused, and it fails yarn typecheck.
3. streaming/index.js — kept glitch-soc's explanatory comment and indentation instead of dev's de-indented variant. No functional difference.
4. app/javascript/{flavours/glitch,}/components/autosuggest_textarea.jsx — dropped a whitespace-only line that dev carries.
5. Formatting — dev's theme SCSS (app/javascript/**/theme/*.scss, modern.scss, modern-urusai-fixes.scss, tangerineui*) was normalised with the repo's own tooling (oxfmt, stylelint --fix) so that yarn format:check and yarn lint:css pass; modern.scss also gained a stylelint-disable-next-line comment for the (invalid, browser-ignored) font-display declaration on body. dev's whitespace-only tweak to config/vite/plugin-sw-locales.ts is gone, because glitch-soc renamed that file to .mts and the built .mts carries glitch-soc's content.
6. Lint/type fixes so the repository's own checks pass. dev widens ESLint coverage by adding files: ['**/*.js', '**/*.jsx', '**/*.ts', '**/*.tsx'] to the block that carries ecmaVersion: 2021, the TypeScript parser and the browser globals, so that the block applies to .jsx files as well. This patch set keeps that coverage — dropping '**/*.jsx' from the list makes .jsx files fall back to a lower ecmaVersion and they fail to parse (Parsing error: Unexpected token =, 14 files) — and fixes the findings instead:
   • app/javascript/flavours/glitch/features/gif_modal/index.tsx — dropped an inline eslint-disable-next-line jsx-a11y/no-autofocus that the rule being off makes redundant (--report-unused-disable-directives).
   • app/javascript/flavours/glitch/features/ui/components/sign_in_banner.jsx, app/javascript/flavours/glitch/features/local_settings/navigation/item/index.jsx and app/javascript/flavours/glitch/features/notifications/components/pill_bar_button.jsx — added type='button' (react/button-has-type); none of them is inside a form, so the default is only made explicit.
   • app/javascript/{flavours/glitch,mastodon}/components/scrollable_list/index.jsx and app/javascript/{flavours/glitch,mastodon}/components/status/legacy/content.jsx — replaced @param {*} / @param {any} with concrete jsdoc types (jsdoc/reject-any-type).
   • eslint.config.mjs — added **/*.mjs to that files list (without it the repository's own config file is parsed at the default ecmaVersion and fails on import.meta), restored glitch-soc's mjs: 'never' (dev had commented it out), and added an override turning import/no-restricted-paths off for app/javascript/flavours/glitch/locales/*.js, which exist to extend the vanilla locale JSON files.
7. app/javascript/mastodon/features/ui/index.jsx — kept glitch-soc's newer forms where dev still has the older ones: the named { BundleColumnError } import (glitch-soc changed that module's export), the one-line if (this.dataTransferIsText(e.dataTransfer)) return; and the trailing return;. Semantically identical to dev's variants.
8. Two runtime fixes for code that glitch-soc has since changed (both are forks of dev's files that break against the current base commit):
   • app/controllers/api/v1/custom_emojis_controller.rb — skip_before_action :require_authenticated_user!, unless: :whitelist_mode? referred to a method glitch-soc renamed to limited_federation_mode? (ApplicationController exports the new name; whitelist_mode? is gone). Every request to /api/v1/custom_emojis raised NoMethodError; the rename restores the fork's intent (custom emojis stay public except in limited-federation mode). Caught by spec/requests/cache_spec.rb.
   • lib/paperclip/qr_decoder.rb — the rescue path called log(...), which Paperclip::Processor does not define, so when the qrtool binary is missing the graceful degradation raised NoMethodError and killed media post-processing (17 media specs). It now uses Rails.logger.warn.
9. Ruby/HAML/locale lint fixes so bin/rubocop, bin/haml-lint and bin/i18n-tasks pass:
   • lib/paperclip/qr_decoder.rb — [a, b].reject(&:blank?) → .compact_blank (Rails/CompactBlank).
   • app/views/admin/settings/other/show.html.haml — the three show_{reblogs,replies}_in_{local,federated}_timelines inputs were 248 characters on one line; they use the repo's multi-line f.input style now (haml-lint LineLength, max 240).
   • config/locales-glitch/en.yml — removed a duplicated notification_mailer.reaction block that dev carries twice, and config/locales-glitch/simple_form.fr.yml — sorted one key (bin/i18n-tasks check-normalized).
10. Fixes to the fork's own specs, all pre-existing in dev:

• spec/requests/api/v1/statuses/reactions_controller_spec.rb — rewritten as a request spec using the real routes (POST /api/v1/statuses/:status_id/react/:id and /unreact/:id); it used to be written as a controller spec (post :create) under spec/requests/, which RSpec types as a request spec, and it called a body_as_json helper that no longer exists. The unreact examples need :inline_jobs because UnreactWorker removes the reaction asynchronously.
• spec/validators/status_reaction_validator_spec.rb, spec/workers/unreact_worker_spec.rb — used bare describe, which spec_helper's disable_monkey_patching! removed; the files could not load at all. Now RSpec.describe.
• spec/services/{react,unreact}_service_spec.rb — the federation examples expected an inline delivery without the :inline_jobs tag the rest of the suite uses.
• spec/policies/status_policy_spec.rb — Fabricate(:react) (no such fabricator; it is :status_reaction) and react.target_account (a StatusReaction has no target_account).
• spec/validators/status_reaction_validator_spec.rb — called status.reactions.build / .create!; Status#reactions is a method returning an Array, the association is status_reactions.
• spec/models/notification_spec.rb — added a reaction notification to the inputs without adding it to the contain_exactly expectation; it now has a reaction_attributes matcher.

11. /api/v1/custom_emojis stays public when unauthenticated API access is disallowed — that is the fork's behaviour, and it is now asserted instead of contradicted: spec/requests/cache_spec.rb (patch 5) expects a successful, non-publicly-cached response for that endpoint under DISALLOW_UNAUTHENTICATED_API_ACCESS=true, where upstream expects an error. The controller's skip_before_action … unless: limited_federation_mode? (deviation 8) is what implements it.
12. Three upstream specs resolve their CSV fixtures explicitly (patch 5): spec/models/form/import_spec.rb and spec/controllers/admin/export_domain_blocks_controller_spec.rb passed bare fixture names to fixture_file_upload, which prefers a path relative to the working directory when one exists — and this repository keeps the operator's domain_blocks.csv at its root (796 rows) next to the 3-row spec/fixtures/files/domain_blocks.csv. They now build the path from file_fixture_path, so the fixture always wins.

Notes and caveats

Behaviour worth knowing before deploying this tree:

• The bubble timeline only ever shows remote accounts. A status is "in the bubble" when its author's domain is in bubble_domains (Status#bubble? → BubbleDomain.in_bubble? → rule_for), and rule_for returns nil for a blank domain, so local accounts can never be in the bubble. Fan-out matches that: broadcast_to_public_streams! publishes locals to timeline:public:local and remotes to timeline:public:remote, timeline:public and — only if @status.bubble? — timeline:public:bubble. The column is labelled accordingly ("posts from people … on other servers selected by {domain}"). A boost of a bubble-domain post is published by the booster's own status, so a local account that boosts a bubble post does not put it on the bubble timeline.
• show_*_in_{local,federated}_timelines defaults to false, and older settings rows are ignored. Upstream Mastodon once used show_reblogs/replies_in_public_timelines; a database migrated through those versions may still hold rows under those keys. They are no longer read. This changes nothing for a fresh install or one that kept the defaults (dev's config/settings.yml also defaulted the old keys to false), but an install whose settings table has the old reblog key set to true used to stream boosts and will stop until the two new toggles are enabled in Administration → Settings.
• Ruby-side checks: run, and green. Ruby 4.0.7, PostgreSQL 18 and Redis are needed (see docs/verification.md for the exact commands); with them, bin/rails db:migrate from an empty database, a db:schema:dump comparison, bin/rubocop, bin/haml-lint, the bin/i18n-tasks set and bin/flatware rspec all run and pass: db:schema.rb is byte-identical to a fresh dump, the linters report nothing, and RSpec is 7934 examples with 0 failures (4 pending, upstream's own). Two ActivityPub::ObjectIntegrityProof ML-DSA examples depend on OpenSSL's post-quantum support and failed in one run and passed in another, so treat them as environment-dependent.

Provenance

How this revision was produced (it is reproducible from the pre-sync dev, branch backup-20261006, and the base commit):

1. git merge of fedi.my.id@dev (b416d280e0) onto b3877d5b245c6fa65f1a7639fa5593c6b6ade8fc (merge-base 64e05b4b2eece29fafeff43c867bb0985c28bc52, the fork's last glitch-soc merge);
2. the 8 conflicts resolved by keeping glitch-soc's newer code and re-applying the fork's intent (flavours/glitch and mastodon components/status/legacy/action_bar/index.jsx, flavours/glitch features/status/components/action_bar.jsx, flavours/glitch and mastodon features/ui/index.jsx, flavours/glitch and mastodon features/ui/components/link_footer.tsx, flavours/glitch features/firehose/index.jsx);
3. the resulting difference folded into the five patch boundaries;
4. the repo's own formatter and linters applied to the files the patch set touches, plus the lint fixes listed under Deviations, and the fixes to the fork's own specs (deviation 10);
5. the patches exported with git format-patch, and the checks in docs/verification.md run — including the Ruby-side checks, which need Ruby 4.0.7, PostgreSQL and Redis (deviation note above);
6. the series applied to glitch-soc b3877d5b24 and merged into the fork as dev (commit 43e2b637c8, tree b473d8b76fe2fb8cd5084e3135a62ea923c2d126, first parent the previous dev b416d280e0, kept as backup-20261006). The push was a fast-forward, and the comparison against the new dev reports the result byte-identical — the numbers in the next paragraph are the ones against the pre-sync fork.

Of the fork's 464 changed files, all 464 survive. Of the 421 files that only the fork touches, 340 are byte-identical to the pre-sync dev and the other 81 are fork-only deviations above (63 theme SCSS plus 18 code and spec files); 9 further deviations are files neither the fork nor glitch-soc changed, which only the patch set edits — the five lint/type fixes, the fan-out spec cases (deviation 1) and the three specs of deviations 11–12. 806 paths in total differ from that pre-sync dev: 620 modified, 77 added and 19 removed by glitch-soc since the fork's last merge, plus those 90 deviations. The fork's own changes are in patches 1–4; patch 5 is the reconciliation layer on top. docs/fork-comparison.md has the full breakdown.