Bug#943905: gnutls28 FTCBFS during guile bindings

Helmut Grohne helmut at subdivi.de
Thu Oct 31 20:44:23 GMT 2019


Hi Andreas,

Thank you for the quick reply.

On Thu, Oct 31, 2019 at 06:54:15PM +0100, Andreas Metzler wrote:
> just a wild guess from looking at the error message: Perhaps
> 
> debian/rules
> if test for crosscompiling
>         export GUILE_AUTO_COMPILE=0
> endif

I didn't attempt this initially, because the suggestion also shows up in
native builds. It didn't appear specific to cross building to me.

I've now checked that it yields the same failure (minus the suggestion).

> Would you mind explaning (or giving a pointer) why you would prefer this
> over using the "noguile" profile?

I'm not opposed to profiles in principle. Indeed, I'm likely one of the
major profile users. There are two reasons for not (over)using them:

1. A profile needs to be maintained. If it doesn't get tested, it likely
   becomes buggy and possibly useless. The difference to "normal" builds
   is a source of bugs on its own. For this reason, I try to minimize
   divergence to normal builds (where feasible).

2. For the current bootstrap testing implementation, every build profile
   means manual work. It means that I need to manually order builds when
   it would work with automatic ordering otherwise. This is a limitation
   of the current test implementation, not a limitation of build
   profiles or how they are supported in the archive.

This also means that I don't mind adding a noguile profile. It just
means that I prefer not to use it if avoidable.

It's a slight preference. At the end of the day, I strongly prefer
things to work in practice.

> I have no strong preference, but /fixing/ generally seems to be a better
> way than disabling selectively.

Ok, let's try that first if you (or someone else) have any other ideas.

Helmut



More information about the Pkg-gnutls-maint mailing list