[pkg-gnupg-maint] Bug#869609: libgpg-error is unecessarily hard to bootstrap for new architectures/ABIs

[Wookey]

Source: libgpg-error
Version: 1.26-2
Severity: wishlist
Tags: upstream

libgpg-error is painful to bootstrap for every new architecture,
because it requires a new header file
src/syscfg/lock-obj-pub.<triplet>.h

This is intended to be generated with a native build of
gen-posix-lock-obj, which is a C program, that looks at libc headers
and generates a header file containing a struct.

This is not practical for an actual new architecture where you don't
yet have a machine/filesystem to run this program on. It is
'correctly' cross-built by the current configury to build a binary for
the HOST arch, but as we want to use it at build-time to generate the
header file this is no use as we want a binary for the BUILD arch,
that would output the correct file for the HOST arch. 

It seems redundant to be running a C program which is just fishing
values out of libc headers to make another header. Just using the
headers directly would remove the need for this bootstrapping. Using
an arch-independent scripting language to generate the header would at
least make bootstrapping new arches a lot simpler.

However, given that the point of all this is to make a locking
interface for users that works on all arches, can we not just get rid
of this whole thing and use the available locking primitives in
linux/bsd/windows directly, thereby making a genuinely opaque interface?

Round about here we reach the edge of my expertise, and I hope Steve
Capper, with whom I discussed this today will chime in with more
details of an actual implementation and you can tell us if there is
some reason why that won't work.

Wookey