Re: failure emails assume nixpkgs changes are to blame

[Ludovic Courtès]

Hi,

Glenn Morris <address@hidden> skribis:

> When a job switches from success to failure, eg
>
> http://lists.gnu.org/archive/html/emacs-buildstatus/2013-11/msg00000.html
> http://lists.gnu.org/archive/html/emacs-buildstatus/2013-11/msg00011.html
>
> there's a footer that says "This is likely due to N commits by X".
>
> Based on the input changes in the first example:
> http://hydra.nixos.org/build/6758185
>
> it seems that it is referring to nixpkgs changes.
> I guess these are unlikely to be the cause of most failures,
> rather it is the changes in the actual package being built (ie, emacs in
> this case) that are almost always (?) to blame.

By design, Hydra knows all the inputs to build processes–in this case an
Emacs checkout and a Nixpkgs checkout (all the packages used during the
build process of Emacs necessarily come from Nixpkgs.)  Thus, it can
tell the exact changes that occurred in Emacs and Nixpkgs between two
builds.

Under the assumption that build processes are deterministic, it just
lists all the committers to Emacs and/or Nixpkgs between the
last-known-good build and the failed build.

My guess is that the Emacs failure above is non-deterministic, or
perhaps due to a transient error (e.g., OOM on the machine that run it),
which makes Hydra’s blame list look awkward.  WDYT?

Thanks,
Ludo’.