* grep open pull requests
@ 2017-01-19 18:12 Jack Bates
2017-01-19 19:02 ` Jeff King
0 siblings, 1 reply; 4+ messages in thread
From: Jack Bates @ 2017-01-19 18:12 UTC (permalink / raw)
To: git
I have a couple questions around grepping among open pull requests.
First, "git for-each-ref --no-merged": When I run the following,
it lists refs/pull/1112/head, even though #1112 was merged in commit
ced4da1. I guess this is because the tip of refs/pull/1112/head is
107fc59, not ced4da1?
This maybe shows my lack of familiarity with Git details,
but superficially the two commits appear identical -- [1] and [2] --
same parent, etc. Nonetheless they have different SHA-1s.
I'm not sure why that is -- I didn't merge the commit --
but regardless, GitHub somehow still connects ced4da1 with #1112.
So my question is, how are they doing that,
and why can't "git for-each-ref --no-merged" likewise
connect ced4da1 with refs/pull/1112/head?
$ git clone \
> --bare \
> --config remote.origin.fetch=+refs/pull/*/head:refs/pull/*/head \
> https://github.com/apache/trafficserver.git
$ cd trafficserver.git
$ git for-each-ref --no-merged
More generally, what I want is to periodically list open pull requests
that add or modify lines containing the string "memset". So far I have
the following in a Makefile. Can you recommend any improvements?
.PHONY: all
all: trafficserver.git
cd trafficserver.git && git fetch
cd trafficserver.git && git for-each-ref --format '%(refname)'
refs/pull --no-merged | \
while read refname; do \
git log -p "master..$$refname" | grep -q '^+.*memset' && echo
"$$refname"; \
done
trafficserver.git:
git clone \
--bare \
--config remote.origin.fetch=+refs/pull/*/head:refs/pull/*/head \
https://github.com/apache/trafficserver.git
Lastly, a question more about GitHub than Git, but: Given the way GitHub
is setup, I hope I can get a list of unmerged pull requests from Git
alone. Can you think of a way to list *open* pull requests,
or is that status only available out of band?
Thanks!
[1]
https://github.com/apache/trafficserver/commit/107fc59104cce2a4b527f04e7ac86695c98b568c
[2]
https://github.com/apache/trafficserver/commit/ced4da13279f834c381925f2ecd1649bfb459e8b
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: grep open pull requests
2017-01-19 18:12 grep open pull requests Jack Bates
@ 2017-01-19 19:02 ` Jeff King
2017-01-19 22:12 ` Jack Bates
0 siblings, 1 reply; 4+ messages in thread
From: Jeff King @ 2017-01-19 19:02 UTC (permalink / raw)
To: Jack Bates; +Cc: git
On Thu, Jan 19, 2017 at 11:12:03AM -0700, Jack Bates wrote:
> I have a couple questions around grepping among open pull requests.
>
> First, "git for-each-ref --no-merged": When I run the following,
> it lists refs/pull/1112/head, even though #1112 was merged in commit
> ced4da1. I guess this is because the tip of refs/pull/1112/head is 107fc59,
> not ced4da1?
Right. Git's `--no-merged` is about commit ancestry.
> This maybe shows my lack of familiarity with Git details,
> but superficially the two commits appear identical -- [1] and [2] --
> same parent, etc. Nonetheless they have different SHA-1s.
> I'm not sure why that is -- I didn't merge the commit --
> but regardless, GitHub somehow still connects ced4da1 with #1112.
The commits differ only in the committer timestamp. Try:
diff -u <(git cat-file commit 107fc5910) \
<(git cat-file commit ced4da132)
Is it possible that somebody cherry-pick or rebased it? Looking at the
history of apache/trafficserver, I don't see any merges, which implies
to me that the project is using a rebase workflow to merge PRs.
It's much trickier to find from the git topology whether a particular
history contains rebased versions of commits. You can look at the
--cherry options to "git log", which use patch-ids to try to equate
commits. Something like:
git for-each-ref --format='%(refname)' 'refs/pull/*/head' |
while read refname; do
if test -z "$(git rev-list --right-only --cherry-pick -1 origin...$refname)
then
echo "$refname: not merged"
fi
done
That's obviously much less efficient than `--no-merged`, but it should
generally work. The exception is if the rebase changed the commit
sufficiently that its patch-id may have changed.
> So my question is, how are they doing that,
I suspect the answer is "somebody clicked the rebase button GitHub"
which simultaneously did the rebase and marked the PR as merged.
> Lastly, a question more about GitHub than Git, but: Given the way GitHub is
> setup, I hope I can get a list of unmerged pull requests from Git alone. Can
> you think of a way to list *open* pull requests,
> or is that status only available out of band?
That information isn't reflected in the git topology. It's in GitHub's
database. You can ask the API:
https://developer.github.com/v3/
There are libraries to help with that:
https://developer.github.com/libraries/
I think that's probably the best answer to your "unmerged" question,
too. Ask the API which PRs are unmerged, and then do whatever git-level
analysis you want based on that.
-Peff
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: grep open pull requests
2017-01-19 19:02 ` Jeff King
@ 2017-01-19 22:12 ` Jack Bates
2017-01-19 22:24 ` Jeff King
0 siblings, 1 reply; 4+ messages in thread
From: Jack Bates @ 2017-01-19 22:12 UTC (permalink / raw)
To: Jeff King; +Cc: git
On 19/01/17 12:02 PM, Jeff King wrote:
> It's much trickier to find from the git topology whether a particular
> history contains rebased versions of commits. You can look at the
> --cherry options to "git log", which use patch-ids to try to equate
> commits. Something like:
>
> git for-each-ref --format='%(refname)' 'refs/pull/*/head' |
> while read refname; do
> if test -z "$(git rev-list --right-only --cherry-pick -1 origin...$refname)
> then
> echo "$refname: not merged"
> fi
> done
>
> That's obviously much less efficient than `--no-merged`, but it should
> generally work. The exception is if the rebase changed the commit
> sufficiently that its patch-id may have changed.
Cool, thanks for all your help! "git log --cherry-pick" works quite
well. One thing: I expected the following to be equivalent, but found
that they're not. Is that by accident or design?
$ git rev-list --cherry-pick --right-only master...refs/pull/1112/head
$ git rev-list --cherry-pick master..refs/pull/1112/head
> I think that's probably the best answer to your "unmerged" question,
> too. Ask the API which PRs are unmerged, and then do whatever git-level
> analysis you want based on that.
Right, that makes sense. Thanks again!
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: grep open pull requests
2017-01-19 22:12 ` Jack Bates
@ 2017-01-19 22:24 ` Jeff King
0 siblings, 0 replies; 4+ messages in thread
From: Jeff King @ 2017-01-19 22:24 UTC (permalink / raw)
To: Jack Bates; +Cc: git
On Thu, Jan 19, 2017 at 03:12:53PM -0700, Jack Bates wrote:
> Cool, thanks for all your help! "git log --cherry-pick" works quite well.
> One thing: I expected the following to be equivalent, but found that they're
> not. Is that by accident or design?
>
> $ git rev-list --cherry-pick --right-only master...refs/pull/1112/head
> $ git rev-list --cherry-pick master..refs/pull/1112/head
It's by design. The "left" and "right" notions are defined only for a
three-dot symmetric difference.
In the first command you've asked git to look at commits on _both_
master and the PR, down to their merge base. It marks the tips with a
"left" and "right" bit, and then those bits propagate down.
In the second command, you've only asked for the PR commits, and there
is no left/right bit at all. So --cherry-pick is doing nothing, as it
has no "left" commits to compare to.
-Peff
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2017-01-19 22:24 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-01-19 18:12 grep open pull requests Jack Bates
2017-01-19 19:02 ` Jeff King
2017-01-19 22:12 ` Jack Bates
2017-01-19 22:24 ` Jeff King
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).