[PATCH] D22463: [RFC] Moving to GitHub Proposal: NOT DECISION!

[Mehdi Amini via cfe-commits]

> On Jul 18, 2016, at 8:23 PM, Tim Northover via cfe-commits <cfe-commits at lists.llvm.org> wrote:
> 
>>> Can't handle the update of the umbrella *because of GitHub*, this could be possible with our own hosting of git for instance.
>>> 
>> Pre-commit hooks are not designed to update the umbrella. Webhooks will be able to update the umbrella with a small external service, as proposed in the IRC.
> 
> I think we could emulate any pre-commit hook we like via GitHub
> WebHooks by having two repositories: llvm and llvm-staging (say).
> 
> People push to llvm-staging, which notifies some LLVM server we own.
> That does basic sanity checks and pushes to llvm proper if passed.

I think that would be terrible in practice, for instance how do you handle situations like:

1) Dev A push commit A
2) Dev B push commit B that changes some lines close to the one changed by commit A
3) sanity check fails on commit A, but llvm-staging contains A and B and can’t get rid of A easily because B would not apply without A.

At this point Dev B gets an email (or other mechanism, I don’t know what you imagined) telling that his changed are rejected for no good reason.

Also reverting to a state "before A” on llvm-staging would mean that that the history would be rewritten and everyone that pulled/fetched in the meantime would suffer .

If we want to go towards pre-check using staging, I believe we should work with pull-request (we’d still have the issue of conflicting PR, but I don’t believe it’d be that bad in practice).
That’d be welcome, but that’d also be a whole other story to setup and maintain!

— 
Mehdi


> 
> It has disadvantages (no instant "success" feedback being the obvious
> one), but would allow us to vet commits with relatively little
> overhead (as James says, running a server responding to webhooks is a
> very different proposition from one hammered by hundreds of developers
> daily).
> 
> I'm not strongly in favour of this, just thought I'd mention it as a
> possibility.
> 
> Tim.
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits