needs-build build records being created when a build isn't actually needed, in obscure corner case

[Max Bowsher]

I was chasing a weird 'failed to upload' entry for happycoders-emacs on the ubuntu ftbfs report, and discovered this oddity:

The only visible build records for happycoders-emacs are "failed to upload" ones for hardy, intrepid, jaunty:
https://edge.launchpad.net/ubuntu/+builds/?build_text=happycoders-emacs&build_state=all

Yet, there's a binary (arch-all) package published in the archive.

It turns out that this originated from an *ia64* hoary build:
https://edge.launchpad.net/ubuntu/jaunty/amd64/happycoders-emacs/2004.08.14

So, in certain rare cases, needs-build build records are being created where the package already exists in the archive, leading to an inevitable upload failure.

[Celso Providelo (cprov)]

RIght, happycoders is arch-indep and should have been built in i386, not ia64. It was probably a problem in the imports we've done from dak, since hoary is pre-Soyuz. The easiest way to fix it is to upload a newer version and let it build properly (on i386), we can also tweak the DB if the NMU isn't possible for jaunty.

[Max Bowsher (maxb)]

As far as I'm aware, the only problems this is causing are a little bit of noise in the FTBFS report, and a needless rebuild every time a new distroseries opens.

Therefore, unless you want to turn this into an enhancement request for Soyuz to not try building versions already in the archive, or you want to work some magic on the DB, I suggest this be sorted via no-change rebuild in the karmic timeframe.

[Celso Providelo (cprov)]

+1 on the no-change source-rebuild.

[William Grant (wgrant)]

On Tue, 2009-04-14 at 09:55 +0000, Celso Providelo wrote:
> +1 on the no-change source-rebuild.
>
> [snip]
>
> ** Changed in: soyuz
> Status: Incomplete => Won't Fix

Isn't it still a bug in gina? Or is gina sufficiently different from
three years ago?

--
William Grant

[Celso Providelo (cprov)]

William,

Nowadays gina uses exactly the same mechanisms for build creation used in general Soyuz process. It would create a i386 build for an arch-indep source. Our tests guarantee that, more over I can't find the code that would have done that.

If, by any chance, it happens again in the future we will know where to look.