I just tested current master and we're working again: checking for flex... flex checking for lex output file root... lex.yy checking for lex library... none needed checking whether yytext is a pointer... yes
Sorted. Is it time that autoconf gets some CI to actually exercise all these paths? It's own 'make check' followed by building a selection of representative software seems like it shouldn't be *too* difficult. Ross On Fri, 17 Jul 2020 at 01:50, Paul Eggert <[email protected]> wrote: > > On 7/16/20 11:40 AM, Zack Weinberg wrote: > > I believe that what's in trunk now should work fine for > > cross-compilation of programs that don't require yywrap > > I thought so too, but after eyeballing the trunk a bit I noticed a shell > portability bug in the recently-added AC_PROG_LEX code. I then found a couple > more instances of the same bug elsewhere in Autoconf (in rarely-executed > code, I > think). > > The problem is that constructs like ${ac_lib:-none needed} don't conform to > POSIX, as POSIX requires that the stuff after the :- be a single shell word. > (The next time I meet Steve Bourne I want to give him an earfull....) > > I installed a doc patch describing the portability pitfall into Savannah > master: > > https://git.savannah.gnu.org/cgit/autoconf.git/commit/?id=2ff883c27e55b0c0d4a448614c1dee0492c9a7da > > and would like to also install the attached patch to fix the code bugs. I have > held off installing this latter patch, though, as I don't want to get in the > way > of whatever testing you're doing.
