Hi Santiago,

we want to remove the package libtool-bin from the archive, because any
attempt of using it breaks cross compilation. The dialog package is a
bit strange in this regard. It's autoconf stuff attempts to detect
whether there is a libtool.m4 and when there isn't attempts to use a
pre-configured libtool (the one from libtool-bin). Unfortunately, last
time it was autoreconf'ed, that happened without libtool.m4. So
basically, making libtool-bin go away here amounts to autoreconfing
dialog after libtoolizing it. And that's pretty much what I did in the
attached patch. The dialog binary and libdialog.la are bit-identical
with this change.

Helmut

hmm - there are two sets of changes - I don't see a reason for the
change to the curses function checks.

(as for libtool - I recall commenting on that, recently)

hmm - there are two sets of changes - I don't see a reason for the
change to the curses function checks.

(as for libtool - I recall commenting on that, recently)

Hi Thomas,

Thank you for reviewing my patch. The curses function check change does
have a reason. It can be solved differently in principle.

When I ran autoheader, config.hin would lack all the defines that should
have come from CF_CURSES_FUNCS while the relevant HAVE_* defines would
still show up in config.log after running configure and therefore the
resulting dlg_config.h would also lack them. That meant that dialog
would perceive a very dysfunctional curses and its shim would fail to
compile. Quite clearly, we shouldn't assume a crippled curses and
config.hin should contain the relevant templates. As it turns out,
autoheader interprets the m4 files and collects the AC_DEFINE and
AC_DEFINE_UNQUOTED invocations, well some of them actually. The
AC_CHECK_FUNCS would be collected whereas CF_CURSES_FUNCS not, even
though both seemed quite similar. The subtle difference is that
AC_CHECK_FUNCS uses AS_FOR (a loop that is evaluated using m4) whereas
CF_CURSES_FUNCS uses a shell for loop. Thus, autoheader would only see a
single, bogus AC_DEFINE_UNQUOTED for all of CF_CURSES_FUNCS and ignore
that. Avoiding this shell loop is key here and I went for manually
unrolling it, because AS_FOR didn't work out initially and unrolling
seemed workable to me. The crucial bit here is that you cannot use shell
for control flow here. If you prefer AS_FOR or some other working
mechanism, that's fine. Just do something about it to avoid dialog
failing to build when we remove libtool-bin from Debian.

Helmut

I don't use autoheader (though it's present in the fork I've maintained for
about the past quarter-century).  The configure script generates the complete
dlg_config.h without that crutch.  Attempting to bypass that will certainly
lead to unnecessary bug reports.

Actually it would be AC_FOREACH, which invokes AH_TEMPLATE

fwiw, CF_CURSES_FUNCS predates that stuff (1997 versus 1999),
and there are other macros which might use those features.

(I added a to-do to follow up on this)

Hi Thomas,

I fear it occurred to me late that I should be using autoconf-dickey
instead of the standard autoconf for dialog. Hence my patch makes it
work the "wrong" autoconf and thus runs autoheader. I see how that would
not be necessary with autoconf-dickey.

Yeah. And if you make dialog work with autoconf-dickey and without
autoheader, then all of this becomes moot anyway.

Feel free to come up with a different solution as long as we stop
relying on /usr/bin/libtool as that's the component that will go away.
We now have one working solution and I'm happy if that is sufficient to
get the ball rolling for a better solution than mine.

Helmut

thanks (on my to-do list)