Deploy keys only work over SSH. A deploy token is their counterpart for HTTPS: a repository scoped credential, used as the password of a Git request, with read or read and write access. It covers Git operations and LFS, and can be regenerated in place.
Signed-off-by: silverwind <me@silverwind.io>
Co-authored-by: Claude Mythos <noreply@anthropic.com>
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: bircni <bircni@icloud.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
With SIGNING_FORMAT = ssh the OpenPGP verification path builds its
GPGSettings from the instance signing key but leaves the format empty,
so it runs `gpg -a --export` on an SSH public key path. Depending on the
local gpg setup that either exports nothing, so an OpenPGP signed commit
reports gpg.error.generate_hash instead of a missing key, or it fails
outright and logs an export error for every such commit.
Both guards are needed. The first covers SIGNING_KEY set to a path with
SIGNING_FORMAT=ssh; the second covers the shipped default
SIGNING_KEY=default, where the format comes from git's own gpg.format
and never gets reconciled with the hardcoded "openpgp". Drop either one
and a working config goes back to broken.
Also raise minimum git version to 2.18 which was already required before this change.
Fixes: https://github.com/go-gitea/gitea/issues/37452
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
Bot co-authors like `dependabot[bot]` render as one long string with the
email inside the name, and never get truncated, so they overflow the
column.
Co-author idents are parsed with `net/mail`, but a git ident isn't an
email address. `mail.ParseAddress` rejects the whole value when the name
holds characters RFC 5322 reserves, like a `[bot]` suffix or a comma, so
the error branch keeps the raw trailer as the display name and throws
the address away. No address means no `mailto:` link, and the anchor is
what `.avatar-stack-names` truncates.
So parse the angle-addr ourselves when `net/mail` won't take it.
Splitting on the last `<` is safe because git strips angle brackets from
idents. The bare-name branch gets the class too.
Fixes https://github.com/go-gitea/gitea/issues/38949
Follow-up to https://github.com/go-gitea/gitea/pull/39068, which
disabled `modernize` entirely.
- re-enable `modernize`, with only the new `embedlit` rule disabled. It
flattens embedded struct literals across ~145 files, and orphans imports
in 6 of them that the fixer does not remove
- apply the rest of the suite: `errors.AsType`, `reflect.TypeAssert`,
`strings.Cut`, and dropping the legacy import comment
- use the new stdlib `uuid` package, `github.com/google/uuid` becomes
indirect
- use `strings.CutLast` in place of manual `LastIndex` slicing in label
scopes, email domains and the diff tree list
- take the header lint skip dirs from the `go.mod` `ignore` directive
and skip dot-directories, instead of hardcoding the list
Assisted-by: Claude Code:claude-opus-5
Only made some necessary changes:
1. remove `GOEXPERIMENT`, only use jsonv2
1. `make fmt`
* `SigningKey` and `Signature` were affected due to some bugs in the
toolchain, so rewrote them
1. remove or fix fragile magic numbers and strings
* the outputs of image/gzip/zlib packages are different
1. update "nolint" comments for the changed lint behaviors
1. add `tls.MLKEM1024`
Gitea keeps completed Actions runs forever. Artifacts and logs expire on
their own schedule, but the run rows never go away, so `action_run` and
its child tables grow without bound.
Adds `RUN_RETENTION_DAYS` to delete completed runs along with their
jobs, tasks and anything the earlier expiries left behind. It defaults
to 400 days, matching how long GitHub keeps run history browsable. A
dedicated `cleanup_action_runs` cron task performs the cleanup, so
admins can schedule it separately from the nightly artifact and log
sweep.
`0` now means "keep forever" for all three retention settings, where
`LOG_RETENTION_DAYS` and `ARTIFACT_RETENTION_DAYS` previously took it
literally and deleted everything at the next sweep.
Docs: https://gitea.com/gitea/docs/pulls/502
----
## ⚠️ BREAKING ⚠️
`RUN_RETENTION_DAYS` defaults to 400, so completed runs older than that
are deleted when the cron task next runs at midnight. Set
`RUN_RETENTION_DAYS = 0` before upgrading to keep all runs.
---------
Co-authored-by: bircni <bircni@icloud.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
Speed up tests: `make test-backend` 103s to 37s, `make test-integration`
908s to 852s.
Most of it is a detached system notice insert blocking on the SQLite
write lock until the busy timeout expired, and `ExternalServiceHTTP`
re-probing on every call with an untimed `http.Get`.
- fixed one correctness bug with nested transactions: files were deleted
while the outer transaction was open, so a later failure could roll the
database back with the files gone
- git push branch counts were far above the hook batch size
- Fix makefile dependencies so running tests and lint work in fresh
worktrees.
---------
Co-authored-by: Giteabot <teabot@gitea.io>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Bound gitignore template selections at both web and API request
boundaries before repository initialization.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
### Description
`NaturalSortCompare` (`modules/base/natural_sort.go`) compares two
numeric run parts by **raw string length**:
```go
if len(part1) != len(part2) {
return len(part1) - len(part2)
}
```
"Longer digit string = larger number" only holds without leading zeros.
With zero-padded numbers the comparison inverts:
- `file0001` vs `file2` → claims `file0001 > file2`, but `1 < 2`
- `a08` vs `a9` → claims `a08 > a9`, but `8 < 9`
This affects any natural-ordered listing where zero-padded and shorter
unpadded numbers mix (branch/tag/file names, etc.).
### Fix
Strip leading zeros before comparing digit-count magnitude; on equal
magnitude fall back to collation, then to the original length so fewer
leading zeros sort first. Added a small `naturalSortTrimZeros` helper
(keeps one char so `"000"` → `"0"`).
Signed-off-by: Seonghyun Hong <s3onghyun.hong@gmail.com>
Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
`published_at` was an alias for `created_at`, so a release created from
an existing tag reported that tag's commit date as its publication time,
and drafts reported one despite never having been published. It is now
stored separately, set when a release is published and null for drafts.
`created_at` in turn means the date of the commit the release points at,
matching what GitHub documents it to be, and the latest release is
selected by it again. Publishing a release for an old commit no longer
takes over the latest badge, and a tag created in the web UI is dated
the same way as one pushed from the CLI.
Fixes https://github.com/go-gitea/gitea/issues/11206
Fixes https://github.com/go-gitea/gitea/issues/38714
Fixes https://github.com/go-gitea/gitea/issues/31789
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Create delegate hook files and directories without group or other write
access, including correcting existing hook directories.
_Assisted-by: Codex:GPT-5_
`FullSteps` only gave the synthetic "Complete job" step the remaining
log range when the last step that had run was also the final step of the
job. A skipped step does not count as having run, so any job ending in a
skipped step left the post step with an empty range: its logs were
stored but never rendered, and the duration showed as `0s`.
Reproducible with any job whose last step is skipped, which is common
for failure notifications:
```yaml
steps:
- run: echo hello
- run: echo never
if: failure()
```
The gate now checks whether the final step is done, which preserves the
behaviour from https://github.com/go-gitea/gitea/pull/29926 of showing
the post step as waiting while steps are still pending.
--> Regression from https://github.com/go-gitea/gitea/pull/29926
---------
Signed-off-by: bircni <bircni@icloud.com>
Workflows using YAML anchors are rejected as invalid, because a workflow
is split into one document per job and an alias whose anchor lands in
another job's document no longer resolves.
Aliases are now expanded once, right after the workflow is parsed and
before anything reads or splits it, bounded like GitHub's parser so
nested aliases cannot expand without limit. Merge keys stay unsupported,
as they are upstream.
Fixes https://github.com/go-gitea/gitea/issues/38983
Signed-off-by: silverwind <me@silverwind.io>
Use "binding:TrimSpace" instead of fragile IsEmptyString
And fix a bug in locale's `HasKey`: it should also try the default
language if current language doesn't have the translation key, a new
test is added.
Better than before, still not good enough (more work can be done in the
future)
And add the missing error handling in the PrivateContext "bind"
middleware.
By the way, picked some "TrimSpace" changes from "fix: trim whitespace
from SMTP address and port - #38934" (fix#38926)
Admin and write team authorize now grant that mode on every unit,
including units added later, instead of only rows present in
`team_unit`. Granular teams keep `authorize=none` and explicit unit
rows.
Closes the `TEAM-UNIT-PERMISSION` design gap from
https://github.com/go-gitea/gitea/pull/34128.
Maybe also fix#15962 (actually maybe it had been fixed before, the root
cause is out-of-sync "access" table)
## Screenshots
only writing selected:
<img width="1399" height="1007" alt="image"
src="https://github.com/user-attachments/assets/1d1b4c49-a59a-47b6-998f-0464a067395b"
/>
_Created with the help of AI_
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Lets users regenerate a personal access token's value in place, keeping
its name and scopes, instead of deleting and recreating it. Useful when
a token was shared with a third party (e.g. an AI agent) and needs to
be invalidated immediately without redoing scope selection.
Follows the same pattern already used for OAuth2 application client
secrets (`GenerateClientSecret`/`RegenerateSecret`).
**Testing**: added a model unit test and a web integration test;
manually
verified in the running dev server that the old token stops
authenticating
and the new one works immediately after regenerating.
<img width="1040" height="245" alt="image"
src="https://github.com/user-attachments/assets/4de0d8b4-1fc4-49cf-a859-95e24d0b2c0a"
/>
Fixes#38683.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
`github.event.inputs` must mirror the raw `workflow_dispatch` payload,
where
GitHub keeps every input as a string. Only the separate `inputs` context
preserves declared types, e.g. booleans. A previous fix coerced boolean
inputs in the single map that fed both contexts, so
`github.event.inputs.someBool` became a real boolean and comparisons
like
`== 'true'` stopped matching.
`github.event.inputs` now stays string-only again. The `inputs` context
used
for server-side `if:` evaluation of needs-gated/matrix-deferred jobs
re-coerces booleans independently, from the job's own workflow
declaration,
so that path keeps working correctly.
Fixes https://github.com/go-gitea/gitea/issues/38896
---------
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
For a `pull_request_target` (PRT) run, Gitea loads the top-level
workflow from the trusted base branch, but any local reusable workflow
it calls (`uses: ./...`) was read from the PR **head** commit, which the
fork author controls.
## Fix
**Record the source commit where the content is read.**
`DetectedWorkflow` now carries a `SourceCommitSHA` filled in next to
`Content`, so the PRT detection pass at the base commit records the base
SHA automatically.
**Defense in depth.** `loadReusableWorkflowSource` pins the PR base
commit for a PRT run's local `uses: ./...` rather than trusting the
stored SHA. This also covers runs recorded before this change, whose
rows still hold the head SHA and would otherwise resolve from the fork
on rerun.
Existing run rows are not migrated.
---------
Co-authored-by: Zettat <zettat123@gmail.com>
This is a prepare and required step for upcoming Gitea Official Mobile
APP which supports login with OAuth2.
The official Gitea mobile app needs the same mechanism. This adds a
builtin application for it:
| | |
|---|---|
| client ID | `b757811a-05c8-4c76-8d74-a5ee3d2073f2` |
| config name | `gitea-app` |
| display name | `Gitea App` |
| redirect URI | `com.gitea.app://oauth/callback` |
Unlike the existing entries, which are CLIs and can therefore use a
loopback `http://127.0.0.1` redirect, a mobile app authorises through a
system browser session (`ASWebAuthenticationSession` on iOS, Custom Tabs
on Android) that can only receive a custom-scheme callback, hence the
custom scheme here.
Bundle downloads run `git bundle create` into a temp file under
`data/tmp/git-repo-content` and copy it to the response, so every
download puts a second, repo-sized copy on disk before the first byte is
sent. If the process dies mid-request that copy is stranded: the startup
sweep only drops files older than 3 days, and cannot remove a directory
that still holds a newer file.
Streaming `git bundle create -` to the response removes that copy.
`CreateArchive` above already uses the same gitcmd pattern, and the
bundle bytes are unchanged (existing integration assertions on the exact
length still pass).
One consequence: git can now fail after output starts, so a mid-stream
failure truncates the body instead of returning an error.
I did not act on the TODO. A temp ref only works under `refs/heads/*`;
with `refs/bundle/temp-*` the bundle carries no branch and clones empty,
so I noted that on the TODO.
Fixes https://github.com/go-gitea/gitea/issues/38447
---------
Co-authored-by: silverwind <me@silverwind.io>
1. add missing CSP header to api & web render endpoints.
2. make jupyter render skip post-processors, nothing to process
3. make ShortLinkProcessor correctly validate URL schemes and respect
the CustomURLSchemes setting
- Hash emails with sha256. Gravatar moved to sha256, and both it and
libravatar.org serve the same image for either hash.
- Drop `strk.kbt.io/projects/go/libravatar` for a 46 line inline SRV
lookup. It could not bound or cancel its DNS query and panicked on an
unexpected resolver error. The replacement carries the request context
and a 3s timeout.
- Fix federated avatars querying DNS for every avatar on every render.
`loadAvatarSetting` compared a cache field that was never assigned, so
each call rebuilt the resolver and dropped its cache. That cache is
gone, both settings are read where they are used.
- Migration 348 recreates `email_hash` with a 64 char hash column and a
`hash_type` column, so a later algorithm change can tell old rows apart.
The MD5 rows are unreachable and their `UNIQUE` email index would reject
the SHA256 replacements.
- Fix a re-saved avatar form replacing an uploaded avatar with a random
one.
- Remove the `duoshuo` `GRAVATAR_SOURCE` alias, that service shut down
in 2017.
- Remove dead i18n key.
Fixes: https://github.com/go-gitea/gitea/issues/34284
Fixes: https://github.com/go-gitea/gitea/issues/28110
Docs: https://gitea.com/gitea/docs/pulls/499
Signed-off-by: silverwind <me@silverwind.io>
Adds `POST helm/api/prov` endpoint for helm repository allowing for upload of provenance files.
Tested manually to a degree but I really didn't want to mess with gpg
again so I'm not sure if helm will correctly verify the chart.
Initial draft made by gemini 3 flash but was finetuned somewhat.
Additionally there's an route that allows for upload of both files via
/api/charts - as separate files in form. If there's any interest in that
I guess it can be added but I think helm is moving to OCI anyway which
we support.
Fixes: https://github.com/go-gitea/gitea/issues/36678
Co-authored-by: silverwind <me@silverwind.io>
Migrate away from this thin ssh wrapper module while adding more test
coverage.
Removes `sessionPartial`, which hand-copied the layout of a private
`gliderlabs/ssh` struct and reinterpreted a pointer to it via
`reflect.UnsafePointer` to reach the permissions of the authenticated
connection. The layout is unchecked, so an upstream field reorder would
mismatch silently.
The builtin server only needs the session channel with `exec` and
`shell`. Serving those on `x/crypto` drops the hack and the dependency,
since `PublicKeyCallback` returns permissions per key and `x/crypto`
assigns them only after verifying the signature.
Two benign behavior changes:
1. Internal session handler errors report exit status 1 rather than 0,
so a client no longer reads a failure as success.
1. An unusable host key is fatal at startup instead of being replaced by
an ephemeral one that would trigger an error at the client.
Enable [`forcetypeassert`](https://github.com/gostaticanalysis/forcetypeassert)
linter to prevent unchecked type assertions. ~650 issues fixed, most
fixes were clean, some use `setting.PanicInDevOrTesting`.
The only behaviour changes are where code would previously send a 500 error
or panic, a 4xx error is now emitted.
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Nested `Package.swift` files overwrote the real package manifest. The
parser matched on the base name and kept the last entry in ZIP order.
`apple/swift-collections` ships `Benchmarks/Package.swift` and
`Utils/Debugger/FormatterFixtures/Package.swift`. The latter sorts after
the root `Package.swift`, so the registry stored a fixture manifest with
the wrong `swift-tools-version`, causing a toolchain mismatch and a
failed build. GRDB, swift-markdown, swift-syntax, sentry-cocoa and
SDWebImage share this layout.
The parser now keeps only manifests from the shallowest directory
holding one. That covers both a package at the archive root and the
single top level directory `swift package archive-source` produces. At
equal depth the first directory by name wins, so an archive always
yields the same metadata.
A nested manifest above the size limit no longer rejects the upload.
Co-authored-by: silverwind <me@silverwind.io>
Pairs with https://gitea.com/gitea/runner/pulls/1143.
Gitea depends on `gitea.com/gitea/runner` for exactly two packages:
`act/model` and `act/exprparser`, the workflow model and the expression
evaluator it needs to parse workflows and to build the task payload the
runner consumes. Pulling the whole runner module in for that is heavy
and puts shared code in the repository of one of the two consumers.
Both packages now live in `gitea.dev/actionslib` (`pkg/model`,
`pkg/exprparser`), the module Gitea and the runner already share for the
runner API, so the dependency on the runner repository is dropped here.
### Changes
- `gitea.com/gitea/runner/act/model` ->
`gitea.dev/actionslib/pkg/model`, `.../act/exprparser` ->
`gitea.dev/actionslib/pkg/exprparser` (22 files, import paths only).
- `routers/api/actions/runner/interceptor.go` takes the `x-runner-uuid`
/ `x-runner-token` names from `gitea.dev/actionslib/pkg/protocol`
instead of repeating the literals the runner also has.
- `go.mod`: `gitea.com/gitea/runner` removed.
---------
Signed-off-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: Zettat123 <zettat123@gmail.com>
1. add correct "refs/heads" prefix to the branch name for commit graph
2. fix incorrect cache key in GetCommitGraphsCount
3. remove the "--" trim for the tag name, there is no security vulnerability, we never do so anywhere else
Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
The arch parser keeps tar member names verbatim. The index writer joins
those values one per line into the pacman database. So a member name
with a newline adds lines to that package's own `files` entry, which
libalpm reads as further fields.
The scope is one package record. An uploader cannot forge entries for
another package, and can set the same fields in `.PKGINFO` anyway. This
is input validation, not a privilege boundary.
`ParsePackage` now drops names that contain CR or LF. `joinFields` drops
such values again when writing the index, which also covers packages
that are already stored. Real packages never carry newlines in file
paths, so well-formed uploads are unaffected.
---------
Co-authored-by: silverwind <me@silverwind.io>
