On 8/22/22 15:58, Zack Weinberg wrote:
There isn’t a bool module right now, are you proposing to add one?
Yes.
I want to get to where _no_ Gnulib module is required for
bool; AC_C_BOOL by itself should be enough.
Yes, that's the long term goal. But in the shorter term Gnulib-using
code can't yet require Autoconf 2.72+ so it will need to use a new
module in Gnulib (maybe it should be called c-bool instead of bool).
This module will merely say "You should use AC_C_BOOL" and will contain
an implementation of AC_C_BOOL for use in Autoconf 2.71-. The substitute
implementation should be a copy of what's in Autoconf 2.72.
Sure, but why can't the non-autoheader case define bool, false, and true
instead of defining C_BOOL_DEFINITION? That will keep the code simpler
for these people.
I don’t understand what you are suggesting here.
My idea is that for the non-autoheader case, where #defines are
implemented via -D options given to the compiler, Autoconf arranges to
pass '-D bool=_Bool -D true=1 -D false=0' to compilers that don't have
either bool or stdbool.h. That way, user code need not ever #define
bool or true or false; non-autoheader user code can say just this:
#if defined C_NEED_STDBOOL_H && !defined __cpluplus
# include <stdbool.h>
#endif
and autoheader user code (the typical case) need not say anything other
than the usual "#include <config.h>".
Admittedly this is only a minor improvement, as I think pretty much
everybody who uses Autoconf and C also uses Autoheader and config.h.
Looking at the existing Gnulib code, though, suggests another possible
reason why we might need the fallback code to `signed char` (or
Gnulib’s enumeration): allegedly C99-compliant compilers whose Boolean
implementation is buggy. My AC_C_BOOL preserves the existing test for
a buggy Boolean type, from AC_CHECK_HEADER_STDBOOL, so, if we do as
Paul suggests and make configure error out when there’s no usable
Boolean, that will hard reject those compilers as well. I’m fine with
that. Are you fine with that?
That should be OK as far as I know; Bruno may have further thoughts.
One other issue comes to mind: bool bitfield in structs. Although
Autoconf currently tests that these compile, it doesn't test that bool
bitfields of width 1 have the same memory layout as unsigned int
bitfields of width 1. Emacs has this issue and already addresses it
(look for "bool_bf" in the Emacs source). I don't think Autoconf needs
to worry about bool bitfield implementation, as (a) the issue comes up
pretty rarely and (b) neither current Autoconf nor current Gnulib
addresses or solves the issue. However, the issue might be worth
documenting in Autoconf.
