MR does not merge

Thu Jan 17 13:12:07 UTC 2019

+1 on Simon's plan. I explicitly agree with the move not to validate before the final commit (if a rebase is necessary); note that the successful rebase is guaranteed not to have any conflicts (or else step 3 would have failed). We should continue to do CI on the commits that have actually landed, so the final result is validated.

And here's another idea to grease the wheels: I imagine the majority of commits have essentially no chance of showing breakage in the variety of platforms/build methods that we have. Right now, we have [skip ci] as a way of telling the infrastructure to skip. What if we had [full ci] requesting that all the different configurations were run before merging? Without [full ci], only the Linux validation would take place; if that's green, merge. Regardless of the setting of [full ci], the full CI matrix would be run after merging. (IIUC, this is different than [skip ci], which also stops CI from happening after the merge.) Note that this new plan requires opt-in when the author is worried about breakage, but would still catch breakage ere long when the full CI matrix is run after merging. In the meantime, even if full CI fails and the patch author is working on fixing it, the rest of us still know that the build succeeds at least on Linux, so we can continue to plow forward. (Except maybe for Simon. Sorry, Simon.)

I don't know if this new idea is necessary, but it might clear logjams if we notice that they happen often.

Richard

> On Jan 17, 2019, at 7:41 AM, Simon Peyton Jones via ghc-devs <ghc-devs at haskell.org> wrote:
> 
> To me it seems clear that GUARANTEEING that a patch won't break the master branch is difficult; it serialises all patches, and it may require many iterations.
> 
> Here's an obvious alternative for the "merge to master" pipeline;
> 
> 1. Rebase the MR on master.  If that fails, send email to the author.
> 
> 2. Full validate.  If that fails, send email to the author.
> 
> 3. Rebase on master again (if master has changed). If that fails,
>   send email to author
> 
> 4. Commit to master
> 
> Note that
> 
> * There is a small possibility that changes between starting (2) and
>  finishing (3) will meant that, even though (3) succeeds cleanly, there
>  is a semantic bug that means that the result fails.
> 
>  But it is extremely unlikely.  And if it does happen, all subsequent
>  builds will fail, so we'll soon know.  Belt-and-braces: you could validate
>  the result of (4) so that blame is always correctly attributed.
> 
> * The process is highly parallel.  You can be validating multiple patches
>  at the same time, in stark contrast to the completely serial nature of
>  our current story.
> 
> Can Gitlab do this?
> 
> Simon
> 
> |  -----Original Message-----
> |  From: Evan Laforge <qdunkan at gmail.com>
> |  Sent: 16 January 2019 15:20
> |  To: Matthew Pickering <matthewtpickering at gmail.com>
> |  Cc: Simon Peyton Jones <simonpj at microsoft.com>; ghc-devs at haskell.org
> |  Subject: Re: MR does not merge
> |  
> |  At work we use "marge bot",
> |  https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.
> |  com%2Fsmarkets%2Fmarge-
> |  bot&data=02%7C01%7Csimonpj%40microsoft.com%7Cb0a1fe00d27e4879ae1e08d6
> |  7bc614f2%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636832488027160974&
> |  amp;sdata=f99fNqu5kgFuFK%2BUEU8d3Z4JDlQek1yYuY2OM9AMFf8%3D&reserved=0
> |  which is an automatic bot that will do the rebase and resubmit thing.
> |  I think it's pretty essential, because otherwise any intervening merge
> |  means you have to babysit the merge button.  It's also more efficient if
> |  you have expensive CI, because it serializes the runs.
> |  
> |  BTW, I've extended marge with a "try rebase, then try merge" strategy,
> |  which is useful if people merge from head, and a "merge CI run" which is
> |  useful if you have an expensive CI you want to run only on merge, not on
> |  every single branch push.
> |  
> |  On Wed, Jan 16, 2019 at 10:56 PM Matthew Pickering
> |  <matthewtpickering at gmail.com> wrote:
> |  >
> |  > There is problem with the interaction between "merge when validated"
> |  > and "fast forward merge only" option.
> |  >
> |  > If anyone commits to master between clicking the button and validation
> |  > finishing then the merge will fail as the patch needs to be rebased
> |  > before it can be merged.
> |  >
> |  > I'm not sure what the plan to deal with this is.
> |  >
> |  > On Wed, Jan 16, 2019 at 2:49 PM Simon Peyton Jones via ghc-devs
> |  > <ghc-devs at haskell.org> wrote:
> |  > >
> |  > > Ben
> |  > >
> |  > > Six days ago I submitted this MR
> |  > >
> |  > > https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgi
> |  > > tlab.haskell.org%2Fghc%2Fghc%2Fmerge_requests%2F109&data=02%7C01
> |  > > %7Csimonpj%40microsoft.com%7Cb0a1fe00d27e4879ae1e08d67bc614f2%7C72f9
> |  > > 88bf86f141af91ab2d7cd011db47%7C1%7C0%7C636832488027160974&sdata=
> |  > > OjHtPXsIpjeW7tPJvheT%2F4hh4aOW7h82bdcifdQ6cfA%3D&reserved=0
> |  > >
> |  > > Just tiny refactorings.  I said “merge when validated”
> |  > >
> |  > > But six days later, it still appears not to have merged.  What’s up?
> |  I was expecting it to merge in a matter of an hour or two.
> |  > >
> |  > > Thanks
> |  > >
> |  > > Simon
> |  > >
> |  > > _______________________________________________
> |  > > ghc-devs mailing list
> |  > > ghc-devs at haskell.org
> |  > > https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fmai
> |  > > l.haskell.org%2Fcgi-bin%2Fmailman%2Flistinfo%2Fghc-devs&data=02%
> |  > > 7C01%7Csimonpj%40microsoft.com%7Cb0a1fe00d27e4879ae1e08d67bc614f2%7C
> |  > > 72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636832488027160974&sd
> |  > > ata=AxazRR%2FEUVEB93FYHAaLig%2FYPBR%2BLH9eAciYs5NTaNU%3D&reserve
> |  > > d=0
> |  > _______________________________________________
> |  > ghc-devs mailing list
> |  > ghc-devs at haskell.org
> |  > https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fmail.
> |  > haskell.org%2Fcgi-bin%2Fmailman%2Flistinfo%2Fghc-devs&data=02%7C01
> |  > %7Csimonpj%40microsoft.com%7Cb0a1fe00d27e4879ae1e08d67bc614f2%7C72f988
> |  > bf86f141af91ab2d7cd011db47%7C1%7C0%7C636832488027160974&sdata=Axaz
> |  > RR%2FEUVEB93FYHAaLig%2FYPBR%2BLH9eAciYs5NTaNU%3D&reserved=0
> _______________________________________________
> ghc-devs mailing list
> ghc-devs at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs