The `ctx.Repo.RefName` was used to be a "short name", it causes a lot of
ambiguity.
This PR does some refactoring and use `RefFullName` to replace the
legacy `RefName`, and simplify RepoAssignment
This adds links to submodules in diffs, similar to the existing link
when viewing a repo at a specific commit. It does this by expanding diff
parsing to recognize changes to submodules, and find the specific refs
that are added, deleted or changed.
Related #25888
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
close#33086
* Add a special value for "SSH_USER" setting: `(DOER_USERNAME)`
* Improve parseRepositoryURL and add tests (now it doesn't have hard
dependency on some setting values)
Many changes are just adding "ctx" and "doer" argument to functions.
By the way, improve app.example.ini, remove all `%(key)s` syntax, it
only makes messy and no user really cares about it.
Document: https://gitea.com/gitea/docs/pulls/138
In history, Gitea could use the system config `/etc/gitconfig` because
some users said that "they might put certNonceSeed in it"
Actually, we shouldn't not use it, because it also causes conflicts
(there are already some fixes like #28848)
To make the system clear, I think it's worth to introduce the breaking
change: add `GIT_CONFIG_NOSYSTEM=1` to all git commands.
## ⚠️ BREAKING ⚠️
For most users, nothing need to do.
If you have made changes to `/etc/gitconfig` to affect Gitea's behavior,
you need to move these config options to Gitea's internal git config
file, it is usually in Gitea's `{AppDataPath}/home/.git` directory.
The latest commit cache is currently used only for listing tree files.
However, a cold start may take longer than directly invoking the Git
command. This PR addresses the issue of slow response times when
accessing raw files, improving performance in such scenarios.
```log
gitea.log:105521:2024/12/23 08:22:18 ...eb/routing/logger.go:68:func1() [W] router: slow GET /xxxx/xxxxxx/raw/commit/xxxxxxxxxxxxxxxxxxxxxxxxxxx/.editorconfig for 172.18.0.5:53252, elapsed 3526.8ms @ repo/download.go:117(repo.SingleDownload)
```
Extract from #32786
`git diff a..b` is equal to `git diff a b` which is different from `git
diff a...b`. For pull request, we should always
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
* Deprecate "gopid" in log, it is not useful and requires very hacky
approach
* Remove "git.Command.SetDescription" because it is not useful and only
makes the logs too flexible
Resolve#31492
The response time for the Pull Requests API has improved significantly,
dropping from over `2000ms` to about `350ms` on my local machine. It's
about `6` times faster.
A key area for further optimization lies in batch-fetching data for
`apiPullRequest.ChangedFiles, apiPullRequest.Additions, and
apiPullRequest.Deletions`.
Tests `TestAPIViewPulls` does exist and new tests added.
- This PR also fixes some bugs in `GetDiff` functions.
- This PR also fixes data inconsistent in test data. For a pull request,
the head branch's reference should be equal to the reference in
`pull/xxx/head`.
In profiling integration tests, I found a couple places where per-test
overhead could be reduced:
* Avoiding disk IO by synchronizing instead of deleting & copying test
Git repository data. This saves ~100ms per test on my machine
* When flushing queues in `PrintCurrentTest`, invoke `FlushWithContext`
in a parallel.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Gitea instance keeps reporting a lot of errors like "LFS SSH transfer connection denied, pure SSH protocol is disabled". When starting debugging the problem, there are more problems found. Try to address most of them:
* avoid unnecessary server side error logs (change `fail()` to not log them)
* figure out the broken tests/user2/lfs.git (added comments)
* avoid `migratePushMirrors` failure when a repository doesn't exist (ignore them)
* avoid "Authorization" (internal&lfs) header conflicts, remove the tricky "swapAuth" and use "X-Gitea-Internal-Auth"
* make internal token comparing constant time (it wasn't a serous problem because in a real world it's nearly impossible to timing-attack the token, but good to fix and backport)
* avoid duplicate routers (introduce AddOwnerRepoGitLFSRoutes)
* avoid "internal (private)" routes using session/web context (they should use private context)
* fix incorrect "path" usages (use "filepath")
* fix incorrect mocked route point handling (need to check func nil correctly)
* split some tests from "git general tests" to "git misc tests" (to keep "git_general_test.go" simple)
Still no correct result for Git LFS SSH tests. So the code is kept there
(`tests/integration/git_lfs_ssh_test.go`) and a FIXME explains the details.
When opening a repository, it will call `ensureValidRepository` and also
`CatFileBatch`. But sometimes these will not be used until repository
closed. So it's a waste of CPU to invoke 3 times git command for every
open repository.
This PR removed all of these from `OpenRepository` but only kept
checking whether the folder exists. When a batch is necessary, the
necessary functions will be invoked.
Fix#31271.
When gogit is enabled, `IsObjectExist` calls
`repo.gogitRepo.ResolveRevision`, which is not correct. It's for
checking references not objects, it could work with commit hash since
it's both a valid reference and a commit object, but it doesn't work
with blob objects.
So it causes #31271 because it reports that all blob objects do not
exist.
Found at
https://github.com/go-gitea/gitea/pull/31790#issuecomment-2272898915
`unit-tests-gogit` never work since the workflow set `TAGS` with
`gogit`, but the Makefile use `TEST_TAGS`.
This PR adds the values of `TAGS` to `TEST_TAGS`, ensuring that setting
`TAGS` is always acceptable and avoiding confusion about which one
should be set.
Fix#31738
When pushing a new branch, the old commit is zero. Most git commands
cannot recognize the zero commit id. To get the changed files in the
push, we need to get the first diverge commit of this branch. In most
situations, we could check commits one by one until one commit is
contained by another branch. Then we will think that commit is the
diverge point.
And in a pre-receive hook, this will be more difficult because all
commits haven't been merged and they actually stored in a temporary
place by git. So we need to bring some envs to let git know the commit
exist.
Running git update-index for every individual file is slow, so add and
remove everything with a single git command.
When such a big commit lands in the default branch, it could cause PR
creation and patch checking for all open PRs to be slow, or time out
entirely. For example, a commit that removes 1383 files was measured to
take more than 60 seconds and timed out. With this change checking took
about a second.
This is related to #27967, though this will not help with commits that
change many lines in few files.
Remove "EncodeSha1", it shouldn't be used as a general purpose hasher
(just like we have removed "EncodeMD5" in #28622)
Rewrite the "time-limited code" related code and write better tests, the
old code doesn't seem quite right.
Misspell 0.5.0 supports passing a csv file to extend the list of
misspellings, so I added some common ones from the codebase. There is at
least one typo in a API response so we need to decided whether to revert
that and then likely remove the dict entry.
Great thanks to @oliverpool for figuring out the problem and proposing a
fix.
Regression of #28138
Incorrect hash causes the user's LFS files get all deleted when running
`doctor fix all`
(by the way, remove unused/non-standard comments)
Co-authored-by: Giteabot <teabot@gitea.io>
Noteable additions:
- `redefines-builtin-id` forbid variable names that shadow go builtins
- `empty-lines` remove unnecessary empty lines that `gofumpt` does not
remove for some reason
- `superfluous-else` eliminate more superfluous `else` branches
Rules are also sorted alphabetically and I cleaned up various parts of
`.golangci.yml`.
To make it more flexible and support SSH signature.
The existing tests are not changed, there are also tests covering
`parseTagRef` which also calls `parsePayloadSignature` now. Add some new
tests to `Test_parseTagData`
Most time, when invoking `git.OpenRepository`, `objectFormat` will not
be used, so it's a waste to invoke commandline to get the object format.
This PR make it a lazy operation, only invoke that when necessary.
Close#29509
Windows, unlike Linux, does not have signal-specified exit codes.
Therefore, we should add a Windows-specific check for Windows. If we
don't do this, the logs will always show a failed status, even though
the command actually works correctly.
If you check the Go source code in exec_windows.go, you will see that it
always returns exit code 1.
The exit code 1 does not exclusively signify a SIGNAL KILL; it can
indicate any issue that occurs when a program fails.
Fixes#29101
Related #29298
Discard all read data to prevent misinterpreting existing data. Some
discard calls were missing in error cases.
---------
Co-authored-by: yp05327 <576951401@qq.com>
Fixes the reason why #29101 is hard to replicate.
Related #29297
Create a repo with a file with minimum size 4097 bytes (I use 10000) and
execute the following code:
```go
gitRepo, err := gitrepo.OpenRepository(db.DefaultContext, <repo>)
assert.NoError(t, err)
commit, err := gitRepo.GetCommit(<sha>)
assert.NoError(t, err)
entry, err := commit.GetTreeEntryByPath(<file>)
assert.NoError(t, err)
b := entry.Blob()
// Create a reader
r, err := b.DataAsync()
assert.NoError(t, err)
defer r.Close()
// Create a second reader
r2, err := b.DataAsync()
assert.NoError(t, err) // Should be no error but is ErrNotExist
defer r2.Close()
```
The problem is the check in `CatFileBatch`:
79217ea63c/modules/git/repo_base_nogogit.go (L81-L87)
`Buffered() > 0` is used to check if there is a "operation" in progress
at the moment. This is a problem because we can't control the internal
buffer in the `bufio.Reader`. The code above demonstrates a sequence
which initiates an operation for which the code thinks there is no
active processing. The second call to `DataAsync()` therefore reuses the
existing instances instead of creating a new batch reader.
Introduce a new function checkGitVersionCompatibility, when the git
version can't be used by Gitea, tell the end users to downgrade or
upgrade. The refactored functions are related to make the code easier to
test.
And simplify the comments for "safe.directory"
---------
Co-authored-by: delvh <dev.lh@web.de>
With this option, it is possible to require a linear commit history with
the following benefits over the next best option `Rebase+fast-forward`:
The original commits continue existing, with the original signatures
continuing to stay valid instead of being rewritten, there is no merge
commit, and reverting commits becomes easier.
Closes#24906
Replace #28849. Thanks to @yp05327 for the looking into the problem.
Fix#28840
The old behavior of newSignatureFromCommitline is not right. The new
parseSignatureFromCommitLine:
1. never fails
2. only accept one format (if there is any other, it could be easily added)
And add some tests.
## Purpose
This is a refactor toward building an abstraction over managing git
repositories.
Afterwards, it does not matter anymore if they are stored on the local
disk or somewhere remote.
## What this PR changes
We used `git.OpenRepository` everywhere previously.
Now, we should split them into two distinct functions:
Firstly, there are temporary repositories which do not change:
```go
git.OpenRepository(ctx, diskPath)
```
Gitea managed repositories having a record in the database in the
`repository` table are moved into the new package `gitrepo`:
```go
gitrepo.OpenRepository(ctx, repo_model.Repo)
```
Why is `repo_model.Repository` the second parameter instead of file
path?
Because then we can easily adapt our repository storage strategy.
The repositories can be stored locally, however, they could just as well
be stored on a remote server.
## Further changes in other PRs
- A Git Command wrapper on package `gitrepo` could be created. i.e.
`NewCommand(ctx, repo_model.Repository, commands...)`. `git.RunOpts{Dir:
repo.RepoPath()}`, the directory should be empty before invoking this
method and it can be filled in the function only. #28940
- Remove the `RepoPath()`/`WikiPath()` functions to reduce the
possibility of mistakes.
---------
Co-authored-by: delvh <dev.lh@web.de>
This should fix https://github.com/go-gitea/gitea/issues/28927
Technically older versions of Git would support this flag as well, but
per https://github.com/go-gitea/gitea/pull/28466 that's the version
where using it (object-format=sha256) left "experimental" state.
`sha1` is (currently) the default, so older clients should be unaffected
in either case.
Signed-off-by: jolheiser <john.olheiser@gmail.com>
When LFS hooks are present in gitea-repositories, operations like git
push for creating a pull request fail. These repositories are not meant
to include LFS files or git push them, that is handled separately. And
so they should not have LFS hooks.
Installing git-lfs on some systems (like Debian Linux) will
automatically set up /etc/gitconfig to create LFS hooks in repositories.
For most git commands in Gitea this is not a problem, either because
they run on a temporary clone or the git command does not create LFS
hooks.
But one case where this happens is git archive for creating repository
archives. To fix that, add a GIT_CONFIG_NOSYSTEM=1 to disable using the
system configuration for that command.
According to a comment, GIT_CONFIG_NOSYSTEM is not used for all git
commands because the system configuration can be intentionally set up
for Gitea to use.
Resolves#19810, #21148
Nowadays, cache will be used on almost everywhere of Gitea and it cannot
be disabled, otherwise some features will become unaviable.
Then I think we can just remove the option for cache enable. That means
cache cannot be disabled.
But of course, we can still use cache configuration to set how should
Gitea use the cache.
The 4 functions are duplicated, especially as interface methods. I think
we just need to keep `MustID` the only one and remove other 3.
```
MustID(b []byte) ObjectID
MustIDFromString(s string) ObjectID
NewID(b []byte) (ObjectID, error)
NewIDFromString(s string) (ObjectID, error)
```
Introduced the new interfrace method `ComputeHash` which will replace
the interface `HasherInterface`. Now we don't need to keep two
interfaces.
Reintroduced `git.NewIDFromString` and `git.MustIDFromString`. The new
function will detect the hash length to decide which objectformat of it.
If it's 40, then it's SHA1. If it's 64, then it's SHA256. This will be
right if the commitID is a full one. So the parameter should be always a
full commit id.
@AdamMajer Please review.
Update golang.org/x/crypto for CVE-2023-48795 and update other packages.
`go-git` is not updated because it needs time to figure out why some
tests fail.
- Remove `ObjectFormatID`
- Remove function `ObjectFormatFromID`.
- Use `Sha1ObjectFormat` directly but not a pointer because it's an
empty struct.
- Store `ObjectFormatName` in `repository` struct
Refactor Hash interfaces and centralize hash function. This will allow
easier introduction of different hash function later on.
This forms the "no-op" part of the SHA256 enablement patch.
The summary string ends up in the database, and (at least) MySQL &
PostgreSQL require valid UTF8 strings.
Fixes#28178
Co-authored-by: Darrin Smart <darrin@filmlight.ltd.uk>
assert.Fail() will continue to execute the code while assert.FailNow()
not. I thought those uses of assert.Fail() should exit immediately.
PS: perhaps it's a good idea to use
[require](https://pkg.go.dev/github.com/stretchr/testify/require)
somewhere because the assert package's default behavior does not exit
when an error occurs, which makes it difficult to find the root error
reason.
This pull request is a minor code cleanup.
From the Go specification (https://go.dev/ref/spec#For_range):
> "1. For a nil slice, the number of iterations is 0."
> "3. If the map is nil, the number of iterations is 0."
`len` returns 0 if the slice or map is nil
(https://pkg.go.dev/builtin#len). Therefore, checking `len(v) > 0`
before a loop is unnecessary.
---
At the time of writing this pull request, there wasn't a lint rule that
catches these issues. The closest I could find is
https://staticcheck.dev/docs/checks/#S103
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
Closes#26329
This PR adds the ability to ignore revisions specified in the
`.git-blame-ignore-revs` file in the root of the repository.
The banner is displayed in this case. I intentionally did not add a UI
way to bypass the ignore file (same behaviour as Github) but you can add
`?bypass-blame-ignore=true` to the url manually.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
From the Go specification:
> "1. For a nil slice, the number of iterations is 0."
https://go.dev/ref/spec#For_range
Therefore, an additional nil check for before the loop is unnecessary.
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
Hi,
We'd like to add merge files files to GetCommitFileStatus fucntions so
API returns the list of all the files associated to a merged pull
request commit, like GitHub API does.
The list of affectedFiles for an API commit is fetched from toCommit()
function in routers/api/v1/repo/commits.go, and API was returning no
file in case of a pull request with no conflict, or just files
associated to the confict resolution, but NOT the full list of merged
files.
This would lead to situations where a CI polling a repo for changes
could miss some file changes due to API returning an empty / partial
list in case of such merged pull requests. (Hope this makes sense :) )
NOTE: I'd like to add a unittest in
integrations/api_repo_git_commits_test.go but failed to understand how
to add my own test bare repo so I can make a test on a merged pull
request commit to check for affectedFiles.
Is there a merged pull request in there that I could use maybe?
Could someone please direct me to the relevant ressources with
informations on how to do that please?
Thanks for your time,
Laurent.
---------
Co-authored-by: Thomas Desveaux <desveaux.thomas@gmail.com>
Close stdout correctly for "git blame", otherwise the failed "git blame"
would case the request hanging forever.
And "os.Stderr" should never (seldom) be used as git command's stderr
Michael Vetter
hiifong
wxiaoguang
Lunny Xiao
TheFox0x7
Rowan Bohde
yp05327
Steffen Schröter
silverwind
Zettat123
Ehsan Shirvanian
techknowlogick
Jason Song
Brecht Van Lommel
Chongyi Zheng
KN4CK3R
coldWater
pengqiseven
charles
Chris Copeland
John Olheiser
Adam Majer
Mihir Joshi
darrinsmart
Nanguan Lin
Eng Zer Jun
Laurent Cahour