Re: [Lilypond-auto] Patchy report from grenouille

[John Mandereau]

Le lundi 06 août 2012 à 10:48 +0200, David Kastrup a écrit :
> This should actually be more diverse.  Before pushing the new
> test-staging, Patchy should redo git fetch.
> 
> It should then check that
> 
> git log -1 origin/staging..test-staging
> 
> has _empty_ output!!!!  If this is _not_ the case, some commits in
> test-staging have been _removed_ from origin/staging.  We've had the
> situation a lot in the past that I tried resetting origin/staging to
> fight bad material getting into master, and then subsequently it was
> still pushed by some already started Patchy.  If we have Patchy
> instances that run for 8 hours, this problem is acerbated.  So Patchy
> should really do a last-minute check that what is going to push is still
> wanted.

Fortunately enough Patchy seldom takes 11 hours like yesterday, but very
frequent runs make this issue critical indeed.


> If it is not empty, abort and send a mail with the message "staging has
> been reset, aborting operation without pushing".
>
> If that's empty, then we can check
> 
> git log -1 origin..test-staging
> 
> If this is empty, pushing will not achieve anything.  Just end without
> sending a mail.

In this last case, it might be good to send an email saying
"origin/staging has a newer revision than test-staging, not pushing", in
case the output of

git log -1 test-staging..origin

is not empty; it would say that an intermediate commit pushed to master
built succesfully, and that's not bad to know (I don't think we should
move commits one by one from staging to master given the current ratio
"available horsepower"/"build time", though).

I'm working on implementing all this.

Best,
John