Re: buildd could run "make -i" twice on failure
-----BEGIN PGP SIGNED MESSAGE-----
On 08/09/13 01:20, Roger Leigh wrote:
> On Sat, Sep 07, 2013 at 11:28:31PM +0200, Daniel Pocock wrote:
>> After a build fails on the buildd, it would be really useful to
>> have the build run again twice with "make -i" and log the output
>> of the second retry
> What's the main advantage of doing so? To log further errors not
> seen due to the first failure? Why not just use "-k" to begin
If -k will try all the same things as -i (which is true for most well
written Makefiles) then it is fine.
I agree that developers could just add -k in the build section of
debian/rules, but the output is quite verbose. One reason I suggest a
re-run of the make with -i (or -k) is to get a more concise summary of
the errors, probably with -j1 as well, and make that available
separately from the normal buildd log output.
> There's no guarantee that "make" is being used for building, and
> it's non-trivial to determine if it is or is not in use, nor how to
> invoke it appropriately. If we want to do it centrally, it would
> be more reliable to add a new target which sbuild or other build
> tools could invoke on error, and which package maintainers could
> add to support their specific requirements.
That would be quite OK. Maybe every target could have an optional "on
As well as a new target (or targets), there would also need to be some
way to have the output of the target redirected to a dedicated log
Later on, this could also lead to more advanced reporting, e.g. rather
than just counting the number of packages that fail on HURD or fail
after some transition, we would be able to count the number of actual
errors and calculate the ratio of errors/package, etc.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)
Comment: Using GnuPG with Icedove - http://www.enigmail.net/
-----END PGP SIGNATURE-----