Bug 256057
| Summary: | ports/ declines multi-branch pushes without ability to override via --push-options | ||
|---|---|---|---|
| Product: | Services | Reporter: | Jan Beich <jbeich> |
| Component: | Core Infrastructure | Assignee: | Mathieu Arnold <mat> |
| Status: | Closed Works As Intended | ||
| Severity: | Affects Only Me | CC: | lwhsu, portmgr |
| Priority: | --- | ||
| Version: | unspecified | ||
| Hardware: | Any | ||
| OS: | Any | ||
I think this is due to the newly deployed ports hook on the server side. It checks if the comment of the cherry-pick’d commit in the quarterly branch has the commit hash in the main branch. I guess because of both commits are in the staging (receiving) area but hook only finds the hash in the final repository causes this. Maybe also checking the staging area can be a solution. Checking in the staging area is more complicated, it would make the hook way much complicated. The workflow is: 1) push to the main branch 2) cherry-pick to quarterly 3) push to quarterly Using --atomic is nice, but if the push to main fails (which it will more often than not) you will have to re-do the cherry-picks anyway. (In reply to Mathieu Arnold from comment #2) Please, provide --push-options=async-cherry or similar. When pushing a security fix the last thing I want to remember is a silly dance around push hooks. No need to replicate how awful Subversion was. For example, when committer != author push (pre-receive) hook suggests how to override: $ git push freebsd Enumerating objects: 5, done. Counting objects: 100% (5/5), done. Delta compression using up to 8 threads Compressing objects: 100% (3/3), done. Writing objects: 100% (3/3), 296 bytes | 296.00 KiB/s, done. Total 3 (delta 2), reused 0 (delta 0), pack-reused 0 remote: remote: ================================================================ remote: jbeich, you are pushing a commit which author and committer are different: remote: remote: author: foo <foo@bar> remote: commit: 987e6796c6daeee9c8516fa623dcd93896bb1421 remote: subject: test remote: remote: Please check the author name and email are correct and then use: remote: git push --push-option=confirm-author remote: ================================================================ To gitrepo-dev.FreeBSD.org:ports.git ! [remote rejected] main -> main (pre-receive hook declined) error: failed to push some refs to 'gitrepo-dev.FreeBSD.org:ports.git' A similar override can be provided for cherry-picks. For consistency it should probably start with "confirm-" prefix. Now, the option name is a bit of bikeshed: confirm-atomic, confirm-atomic-branches, confirm-multi-branch, confirm-multi-cherry, ... confirm-footshooting. We will discuss how to do so, but the complexity will add some more difficulty to it. So, to provide more context, git *-receive hooks gets one `old-hash new-hash ref` at a time on stdin, and they process each ref as they come. If you push multiple refs, there is no guarantee that refs/heads/main would arrive before refs/heads/xxxxQy. To keep things simple and readable, the hooks are stateless. To do what you ask, hooks would need to become stateful, and some, but not all, checks, would have to be deferred to after all refs have been processed, which would defeat the KISS principle into witch they were written. Note that you still can push everything at once, instead of doing this: git push --atomic freebsd main 2021Q2 do this: git push freebsd main; git push freebsd 2021Q2 (In reply to Mathieu Arnold from comment #6) > no guarantee that refs/heads/main would arrive before refs/heads/xxxxQy. How does this affect --push-option? > git push freebsd main; git push freebsd 2021Q2 Duplication like "git push freebsd" makes it error-prone. I've almost pushed all changes from my staging branch by accident. (In reply to Jan Beich from comment #7) > (In reply to Mathieu Arnold from comment #6) > > no guarantee that refs/heads/main would arrive before refs/heads/xxxxQy. > How does this affect --push-option? Well, if we get the quarterly branch before main, we are going to check that the commits exists before we get the commits. > > git push freebsd main; git push freebsd 2021Q2 > Duplication like "git push freebsd" makes it error-prone. I've almost pushed all changes from my staging branch by accident. If your shell does not support history, or you don't trust it, maybe create either a shell or a git alias to do both push at once. |
I often need to push both to main and a quarterly branch. To avoid desync with hashes I use --atomic but this no longer works. If this is intentional to avoid foot-shooting, please, add an escape hatch. $ cd /path/to/ports/latest $ git log --no-abbrev --pretty=fuller @{u}.. commit 1307312bf475a3457d54c8ea2b3ff871519db22a Author: Jan Beich <jbeich@FreeBSD.org> AuthorDate: Fri May 21 10:43:36 2021 +0000 Commit: Jan Beich <jbeich@FreeBSD.org> CommitDate: Fri May 21 15:53:55 2021 +0000 benchmarks/glmark2: backport --fullscreen fix for WAYLAND option $ glmark2-wayland --fullscreen Error: eglCreateWindowSurface failed with error: 0x300b Error: eglCreateWindowSurface failed with error: 0x300b Error: CanvasGeneric: Invalid EGL state Error: main: Could not initialize canvas $ cd /path/to/ports/quarterly $ git cherry-pick -x main $ git log --no-abbrev --pretty=fuller @{u}.. commit 7023f3d11c3c36d5dd0d08a697dff45534f09541 (HEAD -> 2021Q2) Author: Jan Beich <jbeich@FreeBSD.org> AuthorDate: Fri May 21 10:43:36 2021 +0000 Commit: Jan Beich <jbeich@FreeBSD.org> CommitDate: Fri May 21 15:54:30 2021 +0000 benchmarks/glmark2: backport --fullscreen fix for WAYLAND option $ glmark2-wayland --fullscreen Error: eglCreateWindowSurface failed with error: 0x300b Error: eglCreateWindowSurface failed with error: 0x300b Error: CanvasGeneric: Invalid EGL state Error: main: Could not initialize canvas (cherry picked from commit 1307312bf475a3457d54c8ea2b3ff871519db22a) $ git branch -r --contains 1307312bf475a3457d54c8ea2b3ff871519db22a $ git branch --contains 1307312bf475a3457d54c8ea2b3ff871519db22a * main $ git push --atomic freebsd main 2021Q2 Enumerating objects: 33, done. Counting objects: 100% (29/29), done. Delta compression using up to 8 threads Compressing objects: 100% (15/15), done. Writing objects: 100% (15/15), 1.98 KiB | 1.98 MiB/s, done. Total 15 (delta 10), reused 0 (delta 0), pack-reused 0 remote: remote: ================================================================ remote: jbeich, you are pushing a commit to 2021Q2 which is remote: a cherry-pick, but the commit hash is not part of the main branch. remote: remote: This is most probably because you did the cherry-pick before pushing remote: the changes to main and you ended up having to rebase your work, remote: which changed the commit hashes. remote: remote: If you are trying to push both main and 2021Q2 at the remote: same time, you will have to push main first, and then 2021Q2. remote: ================================================================ remote: To gitrepo.FreeBSD.org:ports.git ! [remote rejected] 2021Q2 -> 2021Q2 (pre-receive hook declined) ! [remote rejected] main -> main (pre-receive hook declined) error: failed to push some refs to 'gitrepo.FreeBSD.org:ports.git'