Re: buildd/porter/maintainer roles again
On Tue, Jul 27, 2010 at 05:01:12PM -0500, Peter Samuelson wrote:
> [Wouter Verhelst]
> > Please remember that every time a package fails to function correctly
> > on a particular architecture, barring toolchain bugs, this is a bug
> > in that package itself.
> "Barring toolchain bugs" is a pretty big caveat. Just as big as
> "barring kernel and libc issues", some other reasons for packages
> to fail to build or run on particular architectures.
> There is a perception, which may or may not be grounded in reality,
> that _most_ FTBFS from the Debian buildds are either toolchain, kernel,
> or libc issues. It is certainly my perception.
It's not my perception, but I guess it depends on how you define
"architecture-specific bugs". If by "architecture-specific," you mean
bugs that occur on one set of architectures, but for one reason or
another not the set of architectures that includes the architecture on
which the developer builds and tests his packages, then you're very,
very wrong. If you rather mean bugs that occur on one architecture, and
one architecture only--perhaps with the exception of related
architectures like mips and mipsel, or i386 and amd64--then there's
probably a pretty high amount of toolchain bugs, indeed.
But even then, I doubt it would be true that these bugs are usually
toolchain related. Toolchain bugs are pretty rare; in the 8 or so years
that I've been an m68k porter, there've been only a handfull. There
have, however, been several packages that failed to build on just m68k,
and most of those were because of bugs in the software that only managed
to trip on m68k because of some peculiarities in the hardware.
Confusion between pointer and integer in return values, for instance,
will cause immediate failures on m68k, yet is not a bug in the m68k
toolchain. Similar problems exist on other architectures, and it is my
perception that these happen far more often than toolchain bugs
> And it seems to have been a toolchain issue that started this thread
> (some mips buildd simply cannot not build Java stuff, as I recall).
Yeah, well, java is a bit of a special case, anyway. We've historically
not been able to do much with java on most of our architectures; java
really isn't very portable, even today.
> Do you, as a porter and buildd admin, have a rough idea what percentage
> of FTBFS and arch-specific bugs you see that are ultimately a bug in
> the package, versus an externality like a bad build chroot, bad kernel,
> bad system library, or bad toolchain? If we're talking about 90% vs.
> 10%, for example, that would inform who should really be on the front
> line triaging this stuff.
I don't really keep track of that, I'm afraid.
The biometric identification system at the gates of the CIA headquarters
works because there's a guard with a large gun making sure no one is
trying to fool the system.