`AccessibleRepositoryCondition(user, unitType)` now requires the repo to
have the requested unit enabled and not globally disabled, consistent
with the per-repo permission checks. This affects code search, Actions
run listings, LFS object reuse, recently pushed branches and org project
milestones.
Written by Claude.
The issue reference popup showed its labels without any spacing, and the
other label lists each used their own gap, with wrapped rows spaced
differently from columns.
`.labels-list` is now a wrapping flex list with `--gap-inline` by
default, giving the popup, issue sidebar and project cards equal gaps in
both directions. Lists inline with text, in issue list titles and
timeline events, stay inline so their labels still wrap one by one.
<img width="600" alt="Issue reference popup, issue sidebar and issue
list labels, before and after"
src="https://github.com/user-attachments/assets/fc4f0b9e-2b9d-4060-a2fb-c1bd3b7391db"
/>
The login tab order test fails in WebKit on macOS, so `make test-e2e`
fails locally on every Mac. CI on Linux is unaffected.
- macOS WebKit moves Tab focus only between text fields, plus buttons
and checkboxes when the system keyboard navigation setting is on, and
never to links, like Safari. Option+Tab reaches every element there and
acts as Tab in Linux WebKit, so the test presses Alt+Tab in WebKit.
- Headless WebKit applies the username `autofocus` late, which pulled
focus back from the password field. The test now waits for the autofocus
first, which also covers it.
Follow-up to https://github.com/go-gitea/gitea/pull/40081
Assisted-by: Claude Code:claude-opus-5-5
Remove CSS rules that no longer match anything because earlier changes
removed or restructured the markup they styled.
- Fomantic card styles, unused since the migrate page and profile card
moved off cards
- List item image, table cell and grid column width rules whose last
users are gone
- Dashboard filter menu and actions panel button overrides
Assisted-by: Claude Code:claude-opus-5-5
Since Playwright 1.56, a WebKit click on a native submit button can
return before the form navigation starts, so `toHaveValue` passed on the
old page with the value `fill()` just set, and the next step raced the
reload.
Assert the textarea's text instead, its server-rendered default, which
only the reloaded page has.
---------
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Removing a milestone deadline with `"due_on": null` was untested, only
`""` was. On release/v28 the null case returned HTTP 500
(https://github.com/go-gitea/gitea/pull/40171). main already handles it,
this adds the case to the existing test so it stays that way.
Several code paths skipped checks that their siblings already perform.
- Code comment replies are rejected when they target a review of another
pull request or another user's pending review
- OAuth2 refuses tokens for users who can't sign in, and introspection
reports their tokens and already used refresh tokens as inactive
- AGit proc-receive runs the same pull request permission and wiki
checks as pre-receive
- Repository transfer drops the repository's owner-level scoped workflow
source registrations, and registrations whose source the owner no longer
owns are ignored
- The web editor's base branch copy into a fork and the fork's upstream
divergence info require read access to the base repository
- LFS downloads require sign-in under strict `REQUIRE_SIGNIN_VIEW`, like
git smart HTTP
- Unknown AGit target branches are reported as an unexpected ref again
Copying lines from a diff, file view, blame, code search or code preview
and pasting them into a comment on Gitea or GitHub produced a markdown
table with escaped pipes and stripped indentation. Both use
`@github/paste-markdown`, which converts copied HTML tables unless they
carry `data-paste-markdown-skip`, as GitHub's own code tables do.
Gitea's code tables now set it too.
The test is a bit more complicated because of a upstream bug which
causes `paste-markdown` to insert pasted content twice, once
https://github.com/github/paste-markdown/pull/128 is merged and
released, the negative assertion could be removed.
---------
Signed-off-by: silverwind <me@silverwind.io>
Derive theme colors from a few base colors with `color-mix(in srgb, …)`
and make the colorful palette coherent, so custom themes that import a
Gitea theme can override a base color and get matching shades.
- Every `-light`, `-dark-1` and `-dark-2` shade follows one rule for all
hues, fixing dark orange shades copied from the light theme, black
shades that barely moved and neon tints
- Base colors move at most 4 OKLab units toward a common lightness, the
median in dark and red in light, yellow stays unchanged
- Text, secondary, console, message, surface and alpha colors derive
from their own base, alpha variants now match their names
- Added and removed diff colors share one lightness per level in all
themes, including colorblind ones
- Chart.js and the external render iframe resolve `color-mix()` values
to hex through `getComputedStyle`
- Remove unused `secondary-dark-12`, `-13`, `shadow-opaque` and
`workflow-edge-hover`
Fixes: https://github.com/go-gitea/gitea/issues/30160
<img width="880" alt="Components with the largest color changes, dark
and light, before and after"
src="https://github.com/user-attachments/assets/a9d27636-e3bb-46fa-a3ac-3fec33fd9806"
/>
<img width="880" alt="All theme colors per family, before and after,
unchanged ones dimmed"
src="https://github.com/user-attachments/assets/9977dde3-0749-4926-b110-18c0dc31d632"
/>
---------
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Commit, compare and SHA links render the SHA in code with the rest of
the URL after it, and links to other repositories get an `owner/repo@`
prefix like references do.
- Abbreviated SHAs, branch compares and `<url>` autolinks now render as
commit links
- Pull request commit URLs no longer break into an issue reference plus
stray text
- Only routes that take a SHA are shortened, so file and API URLs stay
plain
Replaces https://github.com/go-gitea/gitea/pull/36725
<img width="620" alt="Commit links before and after"
src="https://github.com/user-attachments/assets/5919c8a0-94e3-4148-ad11-6944bf790572"
/>
Some useful rules from my past experiences, feel free to suggest more.
Dark theme screenshots is subjective of course, but I prefer them.
---------
Signed-off-by: silverwind <me@silverwind.io>
Byte slicing like `t.Description[:255]` can split a multi-byte character
and store invalid UTF-8. Team descriptions were also cut to 255 bytes
although their `MaxSize(255)` validation allows 255 characters.
Team descriptions now truncate by runes. Signup user agents, Actions log
lines and access log request IDs use `util.TruncateStringBytes`, which
now returns short strings untouched.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Several code paths skipped checks their siblings perform, or cleaned up
more than they validated.
- Official team review requests now count, and removing one no longer
marks an unrelated review official
- Release edits only delete files of validated attachments
- Rejected container uploads no longer delete an existing blob
- Attachment removal only removes the uploader's own pending uploads
- OpenID sign-in checks the allow and block lists against the normalized
identifier
- HTTP git access is rejected while a repository is being created, like
SSH
---------
Signed-off-by: silverwind <me@silverwind.io>
Co-authored-by: delvh <dev.lh@web.de>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The Firefox mermaid e2e test fails intermittently with `Clipboard read
operation is not allowed.`, for example in
https://github.com/go-gitea/gitea/actions/runs/41281998298/job/124776072700.
Firefox rejects `readText()` when it overlaps the page's still pending
clipboard write, and `expect.poll` does not retry errors thrown by its
callback, so one overlapping read fails the test.
https://github.com/go-gitea/gitea/pull/39838 only removed the paste
prompt, not this rejection.
- A new `assertClipboardText` helper treats a rejected read as a
mismatch so it is polled again
- The bot token copy test had the same race and uses it too
_Written by Claude (Opus 5.5)._
Pure refactors with no behavior change, mostly reusing existing helpers
and dropping duplicated or dead code. Net −222 lines.
- Reuse existing helpers instead of inlined copies: `getIssueIndex`,
`NewLabel`, `ToCommitUser`, `GetPackageFileDescriptors`,
`util.SplitTrimSpace`
- Merge copy-pasted blocks in review comment deletion, project deletion
and the pre/post-receive hook templates
- Drop dead code: an error `toStringKeys` never returned, impossible
`bytes.Buffer` write errors, redundant nil-map branches, a `base64`
wrapper
- Use stdlib where it fits: `strings.Cut`, `bytes.IndexAny`,
`cmp.Compare`, `min`/`max`, `slices.Concat`, map lookups instead of key
scans
Bump `giteabot` to
[`v1.5.1`](https://github.com/go-gitea/giteabot/releases/tag/v1.5.1) to
label PRs with a `backport/<version>` label that would cherry-pick
cleanly onto the latest release branch with `backport/conflict-free`.
The backport workflow's concurrency group is now per ref, so release
branch pushes can't cancel a pending backport run from `main`.
_Written by Claude (Opus 5.5)._
The `alpine:3.24` base image lags behind fixes published to the Alpine
repository.
Both images now run `apk upgrade` before `apk add`. Packages from `apk
add` are already current, so this only affects the preinstalled base
packages.
_Written by Claude (Opus 5.5)._
The `test-backend#` and `test-integration#` targets rewrote every `.`
into `/`, because GNU make pattern rules never match goals containing
`/`. That broke regexes like `make 'test-backend#TestA.*'`, which
silently selected subtests instead.
Static pattern rules built from the goals accept `/` directly, so the
pattern reaches `go test -run` unchanged. Subtests are now selected like
`make 'test-backend#TestA/sub'`, on GNU make 3.81 and 4.x.
Removes redundant work from the pull request conversation tab.
- Base and head refs resolve through the `cat-file` batch of the git
repos the callers already hold, instead of two `git rev-parse`
processes, reducing the number of git processes from 9 to 7 per page,
about 17 ms faster locally.
- Sidebar reviewers are batch-loaded and code-review comments reuse
their loaded poster, reducing the number of SQL queries from 387 to 358
on a PR with 169 timeline entries and 15 reviews.
Also fixes the commit list of merged pull requests, which resolved the
merge base as a branch name and returned 500.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The issue reference popup built its `/info` request from the raw link
path, so a link ending in `/owner/repo/issues/N` queried an unrelated
location.
It now builds the request from the owner, repo, type and index parsed
from the link, with `parseIssueHref` anchored to the pathname under
`appSubUrl`.
The job emitter resolves a dependency chain in several rounds in memory,
but a job's `if:` reads its needs' results from the database, where a
need skipped or failed earlier in the same pass was still blocked. A job
with `if: always() && needs.build.result == 'skipped'` was therefore
skipped instead of started.
Finished statuses are now persisted at the end of each round, which also
replaces the separate handling of skipped and failed reusable workflow
callers.
The repo-level runner, secret and variable API tests ran as the repo
owner, who bypasses the repo admin check. That's how the regression
fixed in https://github.com/go-gitea/gitea/pull/40174 slipped through.
As suggested in
https://github.com/go-gitea/gitea/pull/40174#issuecomment-6591991824,
they now run as an admin collaborator, which replaces the duplicate
subtest added there.
The tests are also trimmed to one case per status code, because
`services/secrets/validation_test.go` already unit tests name
validation.
Building the backend under QEMU takes 46 to 54 minutes per foreign
platform in the nightly image build, against 7 minutes natively. The
backend stage now runs on the build platform and cross-compiles with
`GOOS`/`GOARCH`/`GOARM`, so every platform builds at native speed.
- `go generate` runs once instead of once per platform.
- The PR docker dry run builds all three platforms in a single job, on
any docker-related change.
---------
Signed-off-by: silverwind <me@silverwind.io>
Render fenced code blocks with Gitea's own code block renderer instead
of `goldmark-highlighting`, using the managed lexer lookup from
`modules/highlight`. Indented and fenced blocks now share one rendering
path.
Rendered output is unchanged. The trailing newline is stripped on copy
in the frontend since https://github.com/go-gitea/gitea/pull/39755.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
WebKit was removed from the e2e tests in
https://github.com/go-gitea/gitea/pull/37315 because Linux WebKit kept
crashing. The crashes came from a use-after-free in the libsoup bundled
with WebKit, which killed the network process and surfaced as `WebKit
encountered an internal error`. macOS WebKit never had them because it
does not use libsoup. Playwright 1.64 bundles libsoup 3.6.6 with the fix
(https://github.com/microsoft/playwright/issues/42803), so WebKit now
runs on all platforms. Builds on
https://github.com/go-gitea/gitea/pull/40140.
The e2e job ran 50 times on `ubuntu-latest` with WebKit enabled, and all
runs passed without WebKit errors. The third browser adds about 1.5 to 2
minutes per run.
1. Playwright installs its system dependencies on GitHub runners too,
which only come with the Chromium and Firefox libraries
1. The passkey test skips Linux WebKit, which has no WebAuthn,
Playwright only stubs `PublicKeyCredential` there
---------
Signed-off-by: silverwind <me@silverwind.io>
Follow-up to https://github.com/go-gitea/gitea/pull/30205, per
https://github.com/go-gitea/gitea/pull/30205#discussion_r4435046605.
Moving a card on a project board is a project operation, so project
write is enough again. Repository boards already show every card to
project writers, who then got a 404 when moving one.
The per-card archived check is removed too. The frontend sends every
card of the target column, so one card from an archived repository
rejected all drops and reorders into that column, silently. Project
workflows already skip archived repositories, and an archived
repository's own board is read-only.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The snap nightly ran on every push to `main`, about 58 runs and 520 job
minutes a week on GitHub-hosted runners, which share the concurrent job
limit with PR CI.
It now runs as a job in `release-nightly.yml`, reusing its 12-hourly
check for new commits on `main`, like
https://github.com/go-gitea/gitea/pull/40138 did for the other
nightlies. A new release tag is still published by the first build after
`main` changes.
Pull requests between branches without a common history show no commits
and can never be merged. The web form already rejects them, but the API
and crafted `..` compare POSTs did not.
The API compared with `..` since
https://github.com/go-gitea/gitea/pull/36105, it now uses `...` again
and returns 422 when there is no merge base.
A failed contributor stats generation never released its lock, so after
the cached error expired, the contributors and code frequency pages kept
saying stats are being generated until a restart. A generation
outlasting the 5 second wait also leaked its goroutine on an unread
channel.
The lock is now claimed atomically and released together with the
completion signal when generation ends, and a cache miss after the wait
counts as still generating.
The starred and watched repository API lists converted repositories
without loading their attributes, so `language` was always empty. They
are now loaded in one batch, owners included.
Three of these handlers also continued after an internal error and wrote
a second response body, they now return.
WebKit was removed from the e2e tests in
https://github.com/go-gitea/gitea/pull/37315 because Linux WebKit kept
crashing. `make test-e2e` on macOS now also runs WebKit, which does not
show these crashes there. CI stays on Linux, because macOS runners would
need about an extra GB of cache storage.
1. The bot account test clicks the user type dropdown, Safari only moves
Tab focus between text fields
1. On macOS, Playwright runs with a single libuv thread, concurrent
`getaddrinfo` lookups of `localhost` there sporadically stall for 5s
---------
Signed-off-by: silverwind <me@silverwind.io>
Building nightlies on every push to `main` and `release/v*` costs about
1000 runner minutes a day, a third of it in runs cancelled by the next
push. Nightlies now run every 12 hours and only build branches with
commits since the last successful run.
---------
Signed-off-by: silverwind <me@silverwind.io>
The post-receive hook wrote each push's new commit into the branch
table. Each push runs its own hook after the ref is updated, so with
pushes in close succession, like CI pushing back or a quick force-push,
a slower first hook can finish last. The table then keeps the older
commit, or marks a re-created branch as deleted.
- The sync reads each pushed branch's state from git under a repo-scoped
lock
- Deletions go through the same sync, so a re-created branch stays
active
- A branch that fails to sync is logged and skipped instead of aborting
the batch
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Giteabot <teabot@gitea.io>
The repository issue and pull request lists filter `access` by
`repo_id`, which no index covers, so every load scans the whole table.
Issue counts on larger repositories merge three single-column indices,
and the API pull request list loads the latest reviews with `id IN
(SELECT max(id) ... GROUP BY ...)`, which MySQL runs as a scan of every
review, about 1.1s on
https://gitea.com/api/v1/repos/gitea/gitea-mcp/pulls?limit=1.
- Index `access.repo_id`
- Index `team_repo.repo_id`
- Replace the single `repo_id`, `is_closed` and `is_pull` indices with
one `issue` index on `(repo_id, is_closed, is_pull)`
- Join the latest review ids instead of `id IN (subquery)`
Median times in ms on MySQL 8.4 with tables sized like gitea.com (1M
access rows, 500k issues, 200k reviews), for a copy of
https://gitea.com/gitea/gitea-mcp and a repository with 9.7k issues:
| Page | Before | After |
|---|---|---|
| `/{owner}/{repo}/issues` | 107 | 40 |
| `/{owner}/{repo}/pulls` | 106 | 32 |
| `/{owner}/{repo}/pulls?state=closed` | 153 | 35 |
| `/{owner}/{repo}/issues/{index}` | 141 | 60 |
| `/{owner}/{repo}/issues`, 9.7k issues | 303 | 53 |
| `/api/v1/repos/{owner}/{repo}/pulls?limit=50` | 180 | 46 |
| `/api/v1/repos/{owner}/{repo}/pulls?limit=1` | 137 | 16 |
| `/api/v1/repos/{owner}/{repo}/issues`, 9.7k issues | 248 | 43 |
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Giteabot <teabot@gitea.io>
A review's text is stored on both the review and its timeline comment,
but editing it in the web UI or through the issue comment API only
updated the comment. The reviews API kept returning the old body
afterwards.
Extracted out of https://github.com/go-gitea/gitea/pull/39865.
*Written by Claude.*
Co-authored-by: Giteabot <teabot@gitea.io>
With the built-in SSH server, closing the client connection cancelled
the `gitea serv` process tree. go-git based clients close it right after
the push report, so post-receive was killed and branches, pull requests,
webhooks and Actions were not updated.
The command now runs under the server's hammer context like with
OpenSSH, where it ends once its pipes close or at shutdown.
Repository-level and instance-level runners still picked up waiting jobs
of repositories with Actions disabled.
https://github.com/go-gitea/gitea/pull/30331 only covered owner-level
runners, by filtering them out in SQL.
Pickup now checks the Actions unit of each job it is about to claim, for
every runner scope, and cancels the job instead when Actions is
disabled, also honoring `DISABLED_REPO_UNITS`. Disabling Actions
therefore cancels its queued jobs once a matching runner polls, instead
of leaving them waiting until `ABANDONED_JOB_TIMEOUT`.
Runs triggered by pull request reviews took their ref and commit from
the review's `commit_id` instead of the pull request. They now use the
pull request head like other pull request events, and their commit
status is set on the commit that ran.
- Let renovate update hyphenated action inputs like `python-version`
- Python 3.14 → 3.15 in CI
- Download gh 2.103.2 directly in the release workflow instead of using
`install-gh-cli-action` with 2.39.1
- Devcontainer Go image 1.26 → 1.27
- Snap build toolchains Go 1.26 → 1.27 and Node 24 → 26
The dashboard counts ran queries whose cost grew with the whole instance
instead of the viewing user. On https://gitea.com/pulls the two review
counts alone take about 1.4s per page load.
- Check the owner of the already joined repository with `EXISTS` instead
of listing every public repository and public user
- Hide superseded review requests with `NOT EXISTS` instead of grouping
every review on the instance
- Skip the review counts on `/issues`, which never shows them
- Index `team_user.uid` for permission and team lookups by user
- Fix indices for table `watch`, `package_property` and `action_run`,
fix#40090
Median page times in ms, before → after, on a synthetic dataset sized
like gitea.com (175k users, 40k repos, 120k issues, 1M comments):
| Page | SQLite | PostgreSQL 17 | MySQL 8.4 | MariaDB 13 |
|---|---|---|---|---|
| `/pulls` | 901 → 56 | 330 → 156 | 189 → 124 | 192 → 218 |
| `/pulls?type=review_requested` | 1797 → 66 | 557 → 254 | 307 → 245 |
604 → 499 |
| `/pulls?type=reviewed_by` | 1646 → 118 | 537 → 316 | 515 → 265 | 528 →
682 |
| `/pulls?type=mentioned` | 1403 → 46 | 426 → 119 | 165 → 116 | 179 →
179 |
| `/issues` | 923 → 33 | 293 → 39 | 178 → 65 | 140 → 41 |
| `/api/v1/repos/issues/search` (anonymous) | 870 → 110 | 144 → 104 |
123 → 87 | 170 → 193 |
MariaDB checks the owner per row on long scans across all public
repositories, which makes a few pages slower there.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Docker Hub refuses anonymous tag pages past 1000 tags, and Renovate then
drops every release date it fetched
(https://github.com/renovatebot/renovate/issues/46925). The 5-day
`minimumReleaseAge` then holds digest updates of `postgres`, `redis` and
`getmeili/meilisearch` back forever. Stopping at 10 pages keeps the
dates of the newest 1000 tags.
Fixes markdown rendering details where Gitea deviated from GitHub
unintentionally, and adds a `/devtest/markdown` page to review markdown
rendering.
- Paragraphs get the intended 1.5 line height
- Blockquotes, alerts and definition lists no longer inherit browser
default margins
- Loose task lists keep their spacing, the checkbox renders inside the
paragraph like on GitHub
- Footnote references render as `[1]` so adjacent ones stay apart
without CSS
- Code in headings, headings in summaries, nested ordered lists, tables,
`<mark>`, alert icons and `<hr>` follow GitHub
Co-authored-by: bircni <bircni@icloud.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Keep notification counts and stopwatches in sync when tabs open or
reconnect, closing the gap behind [this
flake](https://github.com/go-gitea/gitea/actions/runs/31312525323/job/93242429394).
1. Send current notification counts and stopwatch state on each
WebSocket connection, replacing the separate state fetches tabs made on
every connect.
2. Tabs joining an already open connection fetch current state once,
since its snapshot went out before they attached.
3. Send events queued during the handshake before the snapshot, so older
events can't roll it back.
4. Wait for the Redis SUBSCRIBE acknowledgement, so the snapshot read
after subscribing can't miss an event.
Mermaid scales diagrams wider than the markup column down to its width,
but the iframe kept the unscaled viewBox height, leaving a large blank
area below them.
The iframe now uses a CSS `aspect-ratio` capped at the viewBox height,
so it follows the rendered diagram on any width and resize without
measuring, which keeps diagrams in collapsed `<details>` working.
Diagrams that don't scale (`useMaxWidth: false`) keep the fixed height.
Fixes: https://github.com/go-gitea/gitea/issues/40045
- Skip race instrumentation of the transpiled SQLite engine
`modernc.org/sqlite/lib` when `GOTEST_FLAGS` contains `-race`, Gitea
code and the driver boundary stay instrumented. `models/issues` drops
from ~10s to ~5s.
- Hash OAuth2 client secrets with `bcrypt.MinCost` in tests, like user
passwords already use the `dummy` hasher. `TestOAuth2` drops from ~22s
to ~2s on the PostgreSQL job.
- Stop the Redis locker's extension loop directly on `Close` instead of
waiting for its next tick, and run the memory locker tests under
`testing/synctest`. `modules/globallock` drops from ~12s to ~4s.
Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Agents rarely read the `docs` guides when AGENTS.md only asks for
relevant ones. AGENTS.md now names the guide for each area, and rules
agents kept missing go into those guides.
- AGENTS.md maps Go, TS, CSS, Vue, templates, tests, config options,
migrations and refactors to their guide
- The backend guide lists XORM pitfalls behind recent bugs
(https://gitea.com/xorm/xorm/issues/2615): zero-valued struct
conditions, `Cols` and `updated` columns on updates, soft-delete values,
index drops by `SyncAllTables`, MSSQL limits and pagination order
- The development guide requires a `WIP:` companion docs PR copying the
ini text verbatim for `app.example.ini` changes, and says to avoid
migrations unless the alternative is a workaround
Each `release-create-tag` run created a deployment record, which makes
GitHub show a useless "This branch has not been deployed" box on every
PR:
<img width="819" height="113" alt="image"
src="https://github.com/user-attachments/assets/129c8348-1d70-46b9-8996-8872c84a11f2"
/>
With `deployment: false`, the `release-signing` approval and branch
policy still apply.
Replace the unmaintained `chi-middleware/proxy` with an in-tree handler
that detects client IPs correctly behind common reverse proxies.
1. Handle repeated header lines and comma-separated proxy hops.
2. Accept and normalize IPs, including `ip:port`.
3. Keep the client port from the header value or `X-Real-Port` instead
of logging port 0.
4. Prefer the last `X-Real-IP` line.
5. Keep the proxy's address when the selected value is invalid.
6. Read `Forwarded` only when `X-Forwarded-For` is absent.
Docs PR: https://gitea.com/gitea/docs/pulls/560
Fixes: https://github.com/go-gitea/gitea/issues/35682
Fixes: https://github.com/go-gitea/gitea/issues/36814
Co-authored-by: bircni <bircni@icloud.com>
The Packages row in the team settings read "Manage repository
packages.", but packages belong to the owner, so a team's Packages
permission applies to every package of the organization, regardless of
the team's repositories. Reword it to say so. The docs get the same
clarification in https://gitea.com/gitea/docs/pulls/605.
The pnpm `engines` range was still `>= 11.0.0` while `packageManager` is
on pnpm 12. Raise it to match, and set Renovate's range strategy for it
to `bump`, because the default strategy leaves an already satisfied `>=`
range untouched.
Remove the inset bottom shadow from `<kbd>` elements in markup and on
keyboard shortcut hints. The skeuomorphic key cap look is not needed,
the border alone marks the element as a key.
When `navigator.clipboard.readText()` runs before the page's copy has
landed, Firefox shows its "Paste" prompt, which cannot be interacted
with when headless. Avoid that and and use
`dom.events.testing.asyncClipboard` which is a pref meant for testing
that disabes this prompt.
Fixes: https://github.com/go-gitea/gitea/actions/runs/39113819632/job/117635434106
Signed-off-by: silverwind <me@silverwind.io>
`go test -count=2` failed in 12 packages because tests leaked state into
later runs.
- Reset SQLite sequences after each fixture load, like PostgreSQL
already does
- Prepare the database in tests that relied on earlier tests doing it
- Restore mutated globals with `test.MockVariableValue`
`make lint-go` ran `go install` and then a bare `golangci-lint` from
`PATH`, which fails when `GOBIN` is not in `PATH` and overwrites any
`golangci-lint` installed there. It now uses `go run -exec 'env
GOOS=<target>'`, which builds golangci-lint for the host and passes the
lint target `GOOS` only to golangci-lint.
`deps-tools` prebuilds the tools with `go run -exec true` instead of `go
install`, because `go install` does not fill the executable cache that
`go run` uses.
Also drops the unused `TAGS` env and the redundant OS name in
`--build-tags`.
Code comments next to a diff hunk could show one line too high in the
conversation view vs. in the code view that showed them on the correct
line.
- Treat the hunk end in `CutDiffAroundLine` as exclusive, so the line
after a hunk falls back to the unchanged-line context
- Don't count `\ No newline at end of file` markers as lines when
cutting the hunk
- Share the unchanged-line fallback with migrated review comments
Matching git's stderr text is fragile because it breaks silently when
git rewords a message. Where git offers an exit code or a query with a
defined answer, use that instead, and drop matches that could never
trigger. Follow-up to the discussion in https://github.com/go-gitea/gitea/pull/39857.
- Detect a missing remote by `git remote` exit code 2
- Resolve annotated tags to their tree with cat-file `^{tree}` instead
of spawning `ls-tree` when listing tree entries
- Skip the wiki page lookup when the default branch doesn't exist yet
- Remove unreachable "no merge base" and "unknown revision" checks
Hovering a button in a primary or red button group showed a grey frame,
because the generic group hover border applied to colored groups too.
Green groups had no colors at all.
Colored groups now let the hover background fill the border, and green
groups get the same colors as primary and red. The devtest page shows
all three.
<img width="569" height="304" alt="Hovered buttons in uncolored,
primary, red and green button groups, before and after"
src="https://github.com/user-attachments/assets/e0fb6b16-f1f1-459b-9848-9386cd9453c9"
/>
Extracted from https://github.com/go-gitea/gitea/pull/38404
Replace Fomantic's hard-forked jQuery modal with native `<dialog>`
elements, advancing https://github.com/go-gitea/gitea/issues/29849. The
UI is visually unchanged. Dialogs open with `show()` over an `inert`
page, so tooltips, popups and toasts mounted on `<body>` stay usable
above them, unlike with `showModal()`
(https://github.com/whatwg/html/issues/9936).
- Remove the Fomantic modal and dimmer, add `showModal` and `hideModal`
- Declare modals as `<dialog class="modal-dialog">`, shown inside a
`.modal-backdrop`, instead of `ui modal` with its unused size classes
- Remove two modals that could never be opened
- Block the page while a modal is open, close the topmost one on Escape
and restore focus to its opener
- Allow stacked modals, use a confirm modal instead of `window.confirm`
for deleting content history
- Hide popups and toasts when a modal opens or closes
- Keep the release notes modal open when generating notes fails
Fixes: https://github.com/go-gitea/gitea/issues/39643
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The token a request authenticated with now lives on the doer, so
permission and visibility checks apply its scope from the doer alone.
Replaces https://github.com/go-gitea/gitea/pull/39167, following
https://github.com/go-gitea/gitea/pull/39167#issuecomment-6063839619.
1. Public-only tokens get 404 instead of 403 for hidden users, orgs and
repos, like GitHub.
2. Org ownership, site admin, repo and org creation, generation and
transfer endpoints reject public-only tokens.
3. API pull request responses set `head.repo` to `null` and
`head.repo_id` to `-1` when the caller cannot see the head repository.
4. Public-only tokens see org memberships like anonymous callers, so
concealed members stay hidden.
5. Site admin access in activity feeds, follower lists, org members,
Actions run lists and packages needs a session or an `all`-scoped token.
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Several handlers and model helpers skipped checks that their siblings
already perform.
- Only match principal keys when looking up SSH certificate principals
- Return no attachments for an unsaved issue instead of every issue-less
one
- Reject issue and comment attachments when adding release attachments
- Share one permission check for editing, deleting and reacting to issue
content and for comment attachments between the web UI and the API,
refusing them on locked issues for users who can't write
- Reject issue edits, deletion and label changes in archived
repositories via the API
- Apply cross-origin protection to the repository topics endpoint
- Return 404 for the repositories tab of individual users and respect
profile visibility for the heatmap
Fixes: https://github.com/go-gitea/gitea/issues/39226
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: bircni <bircni@icloud.com>
Integration test processes now get their own work directory, config file
and random ports, so several runs can execute in parallel on one
machine. The slowest tests also lose fixed sleeps and needlessly
expensive setup, with their assertions kept.
- Batch PostgreSQL sequence resets, the main per-test fixture cost there
- Cheap hashing params in hash and OAuth2 fixture tests, full-cost
vectors stay
- Single fetch for negative Actions runner assertions instead of 1s
polls
- Queue flushes and backdated timestamps instead of sleeps
- In-memory SQLite test journal, tmpfs for PostgreSQL in CI
- E2E instance writes SSH keys into its work directory instead of
`~/.ssh/authorized_keys`
Remove the misspell spell checker along with its `lint-spell` and
`lint-spell-fix` make targets, its CI step and its custom dictionary.
`make lint` and `make lint-fix` no longer run a spell check.
Co-authored-by: Giteabot <teabot@gitea.io>
https://github.com/go-gitea/gitea/pull/38046 accidentally dropped
`gitea/unescaped-html-literal` from the eslint config, so it has been
off since. Raised in
https://github.com/go-gitea/gitea/pull/39829#issuecomment-6246720235.
- Re-enable the rule, exempting e2e tests like unit tests
- Detect tags anywhere in strings and template chunks, not only at the
start
- Allow `htmlRaw` templates, they escape like `html`
- Fix the existing violations, which escapes the heatmap tooltip and
shortcut keys
The legacy Fomantic `line-height: 1` on buttons made ellipsized button
text clip descenders like `g` letters, like in the PR target branch
selector. Selection dropdowns were 3px taller than the inputs and
buttons next to them because their padding assumed `line-height: 1em`.
- Buttons and selection dropdowns use `--line-height-default` with
smaller vertical padding, `min-height` keeps their sizes for every font
including CJK
- Remove overrides that only compensated for the old values and dropdown
declarations that never applied
- Label-edit and push-mirror modals use regular-size action buttons like
all other modals
Gitea changed the `dist.tarball` it advertises for already published
versions twice:
| Gitea | `dist.tarball` path of `@scope/name@1.0.0` |
|---|---|
| 1.27 | `%40scope%2Fname/-/1.0.0/name-1.0.0.tgz` |
| 28.0, https://github.com/go-gitea/gitea/pull/39061 |
`@scope%2Fname/-/1.0.0/name-1.0.0.tgz` |
| 28.1, https://github.com/go-gitea/gitea/pull/39434 |
`@scope/name/-/name-1.0.0.tgz` |
pnpm 11.5.2+ verifies the tarball URLs pinned in its lockfile against
the registry metadata, so frozen installs of lockfiles written against
1.27 fail with `ERR_PNPM_TARBALL_URL_MISMATCH`.
- Versions uploaded before this PR get their 1.27 URL back.
- New versions store their canonical tarball path at upload, so later
URL changes can't reach published versions. The host still follows the
request.
Lockfiles written against 28.0 for scoped packages, or by pnpm 12 with
`@scope:registry` against 28.1, need those pinned `tarball:` entries
removed, as 1.27 lockfiles of the same versions need the other URL.
Signed-off-by: silverwind <me@silverwind.io>
Since https://github.com/go-gitea/gitea/pull/39703, Gitea fails to start
when its gitconfig holds a `gc.reflogExpire` other than `90` (e.g.
`30.days`), because `git config --unset-all` exits 5 when no value
matches. Value patterns were also unanchored regexps, so `90` also
removed `90.days`.
- Match values exactly in `configUnsetAll` and `configAddNonExist`
- Treat "nothing to unset" as success
- Use `--replace-all` in `configSet` so a key with multiple values no
longer fails startup
Gitea's default `gc.reflogExpire = 90` is parsed by git as
`1990-<month>-<day>`, currently `1990-10-09`, so reachable reflog
entries never expired.
Since git 2.54, auto maintenance runs its reflog-expire task in the
foreground of every push, and its trigger ignores reachability, so busy
repos rerun `git reflog expire --all` on every push without pruning
anything, stalling large repos for over a minute.
- Stop setting `gc.reflogExpire` so git's own defaults apply, and remove
the `90` written by earlier versions
- Treat the legacy `[git.reflog] EXPIRATION` as days, as documented
The next push to each repo prunes the accumulated entries once.
Fixes: https://github.com/go-gitea/gitea/issues/39693
Fix two builds that reported a non-semver version.
Docker images built with a `VERSION` file reported a bare commit hash as
their version, and binaries built from the nightly source tarball
reported `main-nightly`. Docker builds now honor `VERSION` like `make`
does, and the nightly source tarball stores the actual version.
Also remove the obsolete `main` to `main-nightly` rename, the branch
case already appends `-nightly`.
| Build | Before | After |
|---|---|---|
| Docker release | `28.1.0` | `28.1.0` |
| Docker nightly | `29.0.0+dev-83-gab9bcd6a79` |
`29.0.0+dev-83-gab9bcd6a79` |
| Docker with `VERSION` | `c55da3e` | `29.0.0+custom` |
| Release tarball | `28.1.0` | `28.1.0` |
| Nightly tarball | `main-nightly` | `29.0.0+dev-83-gab9bcd6a79` |
Creating a repository with an initial commit, via `auto_init` or a
template with git content, pushed that commit through receive-pack and
three server-side hooks that do nothing for this internal push except
refresh server info. Fetching it into the bare repository instead makes
such repository creation about 3.7x faster locally (median 790ms to
214ms). E2E tests that create initialized repositories get about
0.2-0.3s faster each.
1. Run `update-server-info` after template content generation, which the
push's `post-receive` hook did before.
2. Remove the now unused `InternalPushingEnvironment` and
`PushOptions.LocalRefName`.
Creating a PR fetches the head commit into the base repo, then pushes
the same objects into `refs/pull/N/head`. Since git 2.54, background
repacks race that push:
1. The push can be rejected with "unable to migrate objects to permanent
storage", see
https://github.com/go-gitea/gitea/actions/runs/37171442185/job/111345034829.
2. For heads with 100+ new objects, the repack can delete the reused
pack, leaving the PR ref pointing at missing objects.
Fetching the head commit with `FetchRemoteTempCommit` and setting the PR
ref with `UpdateRef` avoids both, as the fetch transfers nothing when
the objects exist. Fork PR refs are now updated like AGit PR refs
already are, without going through receive hooks.
Also syncs the PR ref when a PR is reopened again.
https://github.com/go-gitea/gitea/pull/37077 inverted that condition, so
a reopened PR kept a stale ref.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: bircni <bircni@icloud.com>
The SSH integration tests read the user's `~/.ssh/config`, so options
like `ControlMaster` reused a connection authenticated with another
test's key and the tests failed with "Cannot find key". Pass `-F none`
so ssh reads no config files.
Rendering a CITATION.cff could use memory far out of proportion to the
file, as every YAML alias copies its target into the formatted citation
and the parser copies `%TAG` prefixes into every node. Files past these
limits show no citation, like unparseable ones do today.
- Skip files over 256 KiB, largest real-world file found is 80 KiB
- Skip files with `%TAG` directives
- Skip files whose aliases add more than 64 Ki nodes and value bytes
- Skip self-referencing anchors, except a sequence listing itself
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Raise the minimum git version to 2.34, the version in Ubuntu 22.04,
Debian 12 and RHEL 8 ship newer, and remove the fallbacks it makes
obsolete.
- Always enable AGit
- Use `diff --skip-to` and `apply -3` unconditionally
- Set the default branch of new repos and wikis via `git init
--initial-branch`
- Detect rebase conflicts via `REBASE_HEAD`
Clean up the navbar template and styling, and fix the navbar stopwatch,
which navigated to the issue instead of opening its popup since
https://github.com/go-gitea/gitea/pull/36965.
1. Add hover background to the create and user menus
2. Simplify navbar HTML and CSS and remove Fomantic styles
3. Render the notification and stopwatch icons once instead of separate
mobile and desktop copies
4. Make the stopwatch a keyboard accessible button whose popup updates
on push and closes when the stopwatch stops
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: bircni <bircni@icloud.com>
`pull_request_review` runs on fork PRs, which includes all backport PRs,
get a read-only token and no secrets, so giteabot cannot write lgtm
labels and statuses there. A no-op `giteabot-review` workflow now
triggers giteabot through `workflow_run`, which gets both. This allows
retiring the legacy fly.io webhook bot.
Part of https://github.com/go-gitea/giteabot/issues/15
MSSQL's default READ COMMITTED makes reads wait on writers, so the
runner pickup deadlocks with concurrent claims, flaking
`TestCreateTaskForRunnerConcurrentClaim`.
- Enable `READ_COMMITTED_SNAPSHOT` on MSSQL so it reads like PostgreSQL
and MySQL
- Read the pickup cursor before claiming, a lost claim could skip
waiting jobs
- Add tests that fail without consistent READ COMMITTED
Performance: Writes on MSSQL now also store the previous row version in
tempdb, the same versioning cost PostgreSQL and MySQL always pay, and
Azure SQL enables it by default. Reads no longer block on writers, and a
32-runner pickup stress test ran 2.5x faster with it.
---------
Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Giteabot <teabot@gitea.io>
Replace citation-js with a Go port of ruby-cff, which GitHub uses for
"Cite this repository", rendering APA and BibTeX server-side and
dropping about 770KB of JS. Output matches GitHub on 1568 of 1571
real-world files, the rest are improvements over GitHub.
- `CITATION.cff` wins over `CITATION.bib`, matched case-insensitively
and through symlinks
- `CITATION.bib` is offered as-is, without APA
Signed-off-by: silverwind <me@silverwind.io>
Vue files only got `eslint-plugin-vue` rules, so stylistic and
TypeScript rules never ran on them, and `import-x` could not see cycles
between `.ts` and `.vue` files.
- Apply the main ESLint config to `.vue` files
- Let `import-x` parse `.vue` files, which surfaced a cycle between
`repo-findfile.ts` and `RepoFileSearch.vue`
- Fix the resulting lint issues
`vue-eslint-parser` is added as a direct dependency because `import-x`
resolves parsers by package name, see
https://github.com/un-ts/eslint-plugin-import-x/issues/381.
The `gitea.com/go-chi/session` package only exists for Gitea, so it
moves into `modules/session` to fix its bugs directly. Fixes the flake
in
https://github.com/go-gitea/gitea/actions/runs/36726154500/job/109923538400.
- Sessions are only written back when changed, so a read-only request
can't revert a concurrent change or restore a logged-out session, like
https://github.com/go-macaron/session/commit/ae808a4a4660c802965c834299ab08f167effd12
- The session cookie is only set once a session holds data
- Every backend refreshes the expiry on load and file sessions are
written atomically
- Also fix https://github.com/go-gitea/gitea/issues/36176
## ⚠️ BREAKING ⚠️
* the `mysql`, `postgres`, `couchbase` and `memcache` session providers
are removed, use `file`, `db` or `redis` instead
* login-related cookies are renamed to `gitea_session` and
`gitea_remember`, if you'd like to use the old names, set `COOKIE_NAME`
and `COOKIE_REMEMBER_NAME` in app.ini
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The `gitea.com/go-chi/cache` package only exists for Gitea, so it moves
into `modules/cache`.
- `ITEM_TTL = -1` disables caching as documented
- Sub-second TTLs round up instead of never expiring
- The Redis adapter no longer grows a `MacaronCache` hash
- Use two-queue cache for in-memory cache
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The `gitea.com/go-chi/captcha` package only exists for Gitea,
so it moves into `modules/imagecaptcha`.
- Reloading an expired challenge shows a new image instead of a broken one
- Every answer is consumed on its first check
- OpenID registration stops after a failed captcha
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
The `gitea.com/go-chi/binding` package only exists for Gitea, so it
moves into `modules/web/binding` to fix its bugs directly. Split out of
https://github.com/go-gitea/gitea/pull/39504.
- GET and HEAD always bind the query
- JSON `null` slice elements and nested `TrimSpace` fields bind
correctly
- Integer fields reject out-of-range values instead of wrapping
- An empty JSON body binds nothing and an unknown binding rule is an
error
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: bircni <bircni@icloud.com>
MariaDB 11.6.2+ defaults `innodb_snapshot_isolation` to `ON`, which
fails REPEATABLE READ transactions with error 1020 when a row they write
changed after their first read. Gitea's background work like push
processing writes the same rows, so merges, issue closes and workflow
runs fail sporadically.
- Use READ COMMITTED on MySQL and MariaDB, like PostgreSQL and MSSQL
- Update xorm to v1.4.3
Replaces: https://github.com/go-gitea/gitea/pull/39494
Fixes: https://github.com/go-gitea/gitea/issues/39492
---------
Signed-off-by: silverwind <me@silverwind.io>
silverwind
