[Pkg-javascript-devel] Bug#1103113: on filing FTBFS due to out-of-memory on i386

[Lucas Nussbaum]

Hi Paul,

On 17/04/25 at 09:12 +0200, Paul Gevers wrote:
> Control: tags -1 moreinfo
> 
> Hi Lucas,
> 
> [Release Team member hat on]
> 
> I always appreciate your QA work on rebuilding Debian, but I'm wondering
> what the value is of filing out-of-memory FTBFS bugs on a 32 bit
> architecture for source packages that only builds arch:all binaries.
> arch:all binaries in Debian are build on 64 bits architectures with more
> memory space than the 32 bits architectures, so I don't think it's worth the
> stress of the maintainers (of arch:all binaries only sources) to look into
> out-of-memory FTBFS RC problems on low address space systems (in this case
> it looks like assumptions in a test, but still). Related, arch:all only
> source packages have no way to avoid you trying to build on i386. Can you
> please share your opinion?
> 
> I haven't demoted the severity of the (currently one) bug I spotted just
> yet, to enable you to respond, but as you can see from my response, I'm
> inclined to do that.

In general: when mass-filing bugs, I try hard to find the right balance
between the time spent filing bugs, and the amount of errors I make.
(Filing bugs later in the release process translates to giving less time to
maintainers to work on them, so there's some value to filing bugs
early, and thus to be efficient at filing bugs.) But yes sometimes I make
mistake, and bugs are reported when they shouldn't. I'm very fine with the
severity being downgraded or discussed. I see bugs as "facts about
packages", and the fact that the test suite fails on i386 exists
independently from the severity assigned to that fact by the release
team. Of course we don't need to track all "facts" in a bug tracker, but
that one sounds useful enough to be tracked.

Regarding the case of arch:all build failures on i386, I think they are
worth reporting to identify that a package that is declared to work on
all architectures does not work on i386 (or does not completely work on
i386). Even if we don't have a great way to translate that to packages
relationships (maybe we should generalize something like Depends:
unsupported-architecture [i386]).

In the specific case of OOMs on 32b architectures, there were only a
handful of bugs so it's probably simpler to address them on a case by
case basis. #1091239 and #1103134 are two other ones, but removing
support for R on 32b architectures is discussed elsewhere.

So, unless you ask me to stop doing so, I will continue to file such
bugs; I will file them as severity:serious by default (unless I can
identify beforehand that they should be filed at a lower severity); I'm
totally fine with the severity being lowered either by the maintainer or
the release team.

Lucas