Re: [bug-gtypist] gtypist fails to build with ncurses[tinfo]

[Tim Marston]

Hi Felix,

On Sat, Jan 11, 2014 at 03:23:38PM +0100, Felix Natter wrote:
> >> There is another problem as well... If I install my own, local copy of
> >> libncursesw in to, say, ~/mylibs/myncursesw and then try to go like
> >> this:
> >>
> >>   $ ./configure "LDFLAGS=-l$HOME/mylibs/myncursesw"
> >>
> >> it will not work.  pkg-config will use the library that the system
> >> knows about instead.  There is an interesting post about it here[1].
> >
> > Ok, good point.
> 
> --> maybe one can use this instead (from ./configure --help, exported by
> PKG_CHECK_MODULES):
> 
> [...]
> 
> ---> so ./configure "NCURSESW_LIBS=-l$HOME/mylibs/myncursesw" should
> work (in conjuction with [1])?

Yes, that would work (i.e., it is still *possible* to use your own
ncurses library during the build).  But it's not a very satisfactory
solution, IMO.  The user *should* be able to set their own library path
with LDFLAGS.

> >> The same post also suggests a way to use PKG_CHECK_MODULES, which we
> >> discussed before, that is "correct":
> >>
> >>   1. invoke AC_CHECK_LIBS *first*, so that the "usual" mechanisms are
> >>   given a chance to work.
> >>
> >>   2. if that fails, invoke PKG_CHECK_MODULES
> >>
> >>   3. if that succeeds, invoke AC_CHECK_LIBS again, in case pkg-config
> >>   returned something invalid.
> >>
> >> I agree with you, though -- step 3 probably isn't too important.  But
> >> would you mind changing your fix to run AC_CHECK_LIBS first, and then
> >> PKG_CHECK_MODULES on failure?  I think that would be a better
> >> solution.
> 
> The problem is that for gentoo
> (https://bugs.gentoo.org/show_bug.cgi?id=483778,
> https://bugs.gentoo.org/show_bug.cgi?id=457530) we need to (optionally)
> add -ltinfo, so we probably could do something like (untested)
> (and the build will fail again if another -lfoo is added to ncursesw.pc):
> 
> AC_CHECK_HEADER(ncursesw/ncurses.h, HAVE_NCURSESW_H=1)
> AC_CHECK_LIB(ncursesw, add_wch, HAVE_LIBNCURSESW=1)
> AC_CHECK_LIB(tinfo, cbreak, [LIBS="-ltinfo $LIBS"]) # for gentoo
> if test -n "$HAVE_NCURSESW_H" -a -n "$HAVE_LIBNCURSESW";  then
>    LIBS="-lncursesw $LIBS"
> else
>    PKG_CHECK_MODULES(NCURSESW, ncursesw, [
>       LIBS="$NCURSESW_LIBS $LIBS"
>       CFLAGS="$NCURSESW_CFLAGS $CFLAGS"], [
>    echo -e "Error:  both library and header files for the ncursesw library\n"\
>     "are required to build this package.  See INSTALL file for"\
>     "further information. On Debian/Ubuntu you need to install 
> libncursesw5-dev."
>    exit 1])
> fi

But why not use AC_SEARCH_LIBS to find the library that the "cbreak"
function is in?  For example (untested):

        AC_CHECK_HEADER(ncursesw/ncurses.h, HAVE_NCURSESW_H=1)
        AC_SEARCH_LIBS(ncursesw, add_wch, HAVE_LIBNCURSESW=1)
        AC_SEARCH_LIBS([ncursesw tinfo], cbreak, HAVE_LIBTINFO_H=1, [
                AC_MSG_ERROR([can't find cbreak() anywhere!])
        ])

Surely that will include libtinfo only as needed (e.g., on gentoo, but
not on distributions that include cbreak in libncursesw, like debian).

To be honest, I don't understand why you want to run pkgconfig at all!
I realise that Gentoo have suggested it as a solution, but I don't see
the advantage in using it over checking for cbreak in libtinfo
ourselves.  Can you explain what advantage using pkgconfig actually
gives us?  (I'm not an autotools expert, so I also might be missing
something!  :o)

Kind regards,

-- 
Tim Marston
ed.am