Fix parent update #419
Fix parent update #419
Conversation
As a result at the followup pull 1. it resamples all the commits from the beginning of times in 'subrepo branch' 2. the above causes merging issue. The fix: remove an extra checking in the update-gitrepo-file while creating the parent currentl failing the tsts: test/branch-rev-list-one-path.t (Wstat: 0 Tests: 7 Failed: 1) Failed test: 7 test/error.t (Wstat: 0 Tests: 17 Failed: 1) Failed test: 17 Parse errors: No plan found in TAP output test/pull-merge.t (Wstat: 0 Tests: 13 Failed: 1) Failed test: 8 test/push-after-init.t (Wstat: 0 Tests: 0 Failed: 0) Parse errors: No plan found in TAP output test/push-new-branch.t (Wstat: 0 Tests: 0 Failed: 0) Parse errors: No plan found in TAP output test/push.t (Wstat: 0 Tests: 17 Failed: 2) Failed tests: 3-4
…ed rev order in rev-list (at least in git 2.22). the fix: add --topo-order qualifier to the rev-list to make the order to be stable across different test. still failing tests: test/error.t (Wstat: 0 Tests: 17 Failed: 1) Failed test: 17 Parse errors: No plan found in TAP output test/pull-merge.t (Wstat: 0 Tests: 13 Failed: 1) Failed test: 8 test/push-after-init.t (Wstat: 0 Tests: 0 Failed: 0) Parse errors: No plan found in TAP output test/push-new-branch.t (Wstat: 0 Tests: 0 Failed: 0) Parse errors: No plan found in TAP output test/push.t (Wstat: 0 Tests: 17 Failed: 2) Failed tests: 3-4
The reason: there is no '.git' directory, just a .git ref. fix: replace 'cd .dir' with 'cd $(git rev-parse --git-dir)'
…the subrepo pull. I'd like this behavior to change. The parent should *change* after the pull.
test/push-after-init.t test/push-new-branch.t the reason: git 2.21 changed the message from Couldn't to couldn't fix: fixed regex to catch both cases.
With the new scheme the pull will set up a new parent for the subrepo. This causes all history before the parent to be truncated for the consequent push. The way around it is to introduce yet another 'push-parent' in the .gitrepo file. The push-parent is set up by the 'clone' and 'push' commands. The push uses it as subrepo_parent. This way all tests passed.
|
I know this is an old PR (and appears to be inactive), but the parent check is affecting a subrepo I've got where the |
|
The reason the parent SHA isn't being updated is that you have changes that are on the local version that haven't been pushed upstream. Therefore the only safe thing to happen is for us to replay the changes and try to integrate the pull into them. I have found that I get a lot more merge failures when using the rebase method than the merge method. Unfortunately that particular patch is unlikely to ever be merged. The section I have marked for 4.2 is patch |
|
Your reply confuses me. I am pushing changes upstream (though not via I can't use
Instead, I |
|
If you think my problem is out of scope for this PR or wouldn't be fixed by it, then I can open a new issue. I only commented here because I thought this PR was open, could be merged, and might have fixed my issue. |
|
This PR would work for your specific workflow, but it breaks a number of others. I would prefer you open another ticket so it would be easier to find the discussion. Unfortunately there isn't currently a great way to handle your issue and I think it is going to require some discussion. |
|
Sure, I'll open another issue with my usecase. I did think up a workaround for now; run EDIT: Not exactly as above, but it's close. Probably need to pull, see if there are still any differences, then reclone if safe (similarly to what I do now). |
|
Hello @jakebailey, |
|
Sure, it's roughly this: $ git subrepo clean <name>
# Split out the changes still applied to the subrepo but not upstreamed.
$ git subrepo branch <name>
$ pushd .git/tmp/subrepo/<name>
# Diff the worktree with upstream.
$ git diff --quiet <new hash of subrepo>
# If this succeeds, then:
$ popd
$ new_parent=$(git rev-parse HEAD)
$ git config --file path/to/.gitrepo subrepo.parent $new_parent
# Shift up the parent commit to simulate a forced reclone.
$ git commit -am "Update git-subrepo parent"But, in a script. Alternatively if the previous commit was a |
I had found that the parent pointer does not change after the pull. As a result, the next pull does re-sampling of old commits and ends up in merge conflicts in some cases. To fix the issue, i commented out the extra check in the update-gitrepo-file (# if [[ $upstream_head_commit == $output ]]; then).
This works in all my cases. It also works in all tests but "pull-merge.t". The test did sample the parent before pull. So, i modified it to sample the parent after the pull.
There was also an issue with 'push.t'. The push after pull had truncated history, because the parent pointer had been moved before the push is done. The solution for the issue was to add yet another parent pointer as 'push-parent'. So, the .gitrepo has another entry now.
On the way i had fixed a few issues:
git 2.21 changed word casing in the 'Cc]ouldn't find remote ref' message. I fixed the regex
branch... tests were failing sporadically at least with git 2.22. The order of transactions reported by rev-list in presence of branches was different from test to test. looks like --topo-order fixed it.
error.t was failing inside a submodule, because there was a ref to '.dir'. So, the 'cd' command failed. I fixed the test.