Re: libidn malloca compile fails GNULIBHEADERS_OVERRIDE_WINT_T not set

[Brian Inglis]

On 2021-07-26 15:59, Brian Inglis wrote:
> On 2021-07-26 12:17, Simon Josefsson wrote:
> > Brian Inglis writes:
> > > Under Cygwin 64 and 32, libidn malloca compile fails because gl/stdint.h
> > > is generated (for some reason, possibly because gnulib is yet to be or
> > > may not be ported to Cygwin, as Cygwin uses newlib rather than glibc,
> > > and is not Linux) with no value set for GNULIBHEADERS_OVERRIDE_WINT_T
> > > in #if, as no tests are configured to set the value under Cygwin:
> > > please see attached log for details.
> >
> > Thanks for the report.  This can happen if an old wint_t.m4 is used.
> > Did you build directly from the tarball?  Did you run any of autoreconf,
> > autopoint or libtoolize?  Make sure wint_t.m4 in your build tree is
> > 'serial 11'.  See some comments in bootstrap.conf..
>
> Yes - built using cygport CygWin ports utility (based on Gentoo portage 
> with bits of RPM spec added, as RedHat volunteers and Fedora users are 
> core distro and package maintainers) using bash script variable 
> definitions and high level function class overrides, from GNU mirrors 
> tarball, running autoreconf, autoconf, automake.
> As libidn uses gettext and Cygwin uses DLLs not .so shared libs, 
> autopoint, libtoolize, autoconf, autoheader, and automake are run by 
> autoreconf.
>
> > I don't have access to a cygwin machine so I can't easily reproduce this
> > myself.  One thing you can check is the ./configure output: does it
> > contain 'whether wint_t is large enough'?  Otherwise something rebuilt
> > the ./configure script for you and used the wrong wint_t.m4 file.
>
> As far as I can find, the only wint_t.m4 is in your gl/m4 directory.
> But from the config log, it does not appear to be invoked, as there is 
> no "large enough" test.
>
> Thinking there might be issues from the previous build, I blew that 
> whole build tree from yesterday for 1.38 away and reran, and now the 
> large enough test and result appears!?
>
> So something must have glitched somewhere during an earlier attempt, 
> causing a later attempt to produce that error, although exactly the same 
> thing happened in my earlier attempts to build current versions!?
>
> Thanks very much for all your help. Sorry if I wasted your time. I'll 
> have to be more careful with these builds, to ensure if anything goes 
> wrong, everything gets wiped before retries.
>
> However, success means I can at last move on to doing cross-builds of 
> mingw64 64 and 32 bit releases, and if all succeeds, adopt all Cygwin 
> libidn packages, which were orphaned when the previous maintainer got 
> overloaded after the IBM acquisition of RedHat, and update all releases 
> to current from 1.33!

Aargh!
It's back and I can now see lib/wint_t.m4 downgraded to serial 5 while 
gl/wint_t.m4 retained as serial 11!

It looks like it's picking that up from:

        ... 1053 Oct 23 2016 /usr/share/aclocal/wint_t.m4

which appears to be from Cygwin's latest gettext-devel release 0.19.8 so 
outdated, but yours is being detected by autoreconf as 0.19.3 so it 
thinks an "upgrade" is needed.

I'm having problems finding a current wint_t.m4 in gettext.git on savannah.

Could you please check your current build release of gettext-dev/devel, 
see where it's sourced from, where it's built, and if it's got any local 
patches for wint_t.m4?

And I'll dig into where autopoint is checking your gettext release, to 
see where the problems lie.

Then I'll see about upgrading and adopting gettext for Cygwin! ;^>

--
Take care. Thanks, Brian Inglis, Calgary, Alberta, Canada

This email may be disturbing to some readers as it contains
too much technical detail. Reader discretion is advised.
[Data in binary units and prefixes, physical quantities in SI.]