[bug-gnulib] Re: automatic prototypes

[Simon Josefsson]

Bruno Haible <address@hidden> writes:

> Simon Josefsson wrote:
>> It seems safe to assume that if a
>> gnulib module A depend on another module B, the AH_BOTTOM from module
>> B should come before the one for A.
>
> I don't see a relation between the order in the config.h file and the
> module dependencies.

Not even for the order of AH_BOTTOM statements?  If a module A depend
on module B, the AH_BOTTOM statement of module A may need to assume
that the AH_BOTTOM statement from module B has been evaluated.  There
is some relation to the dependency order here.

However, I now realize the sorting must be (I believe) a strict
topological sort.  Stand-alone modules that provide C89/C99/POSIX
types that other modules implicitly assume must have their AH_BOTTOM
evaluated before the AH_BOTTOM statements from the other modules are
invoked.

For example, consider a module A that provide ptrdiff_t.  Module B may
need ptrdiff_t in the header file, i.e., in the AH_BOTTOM statement,
but should probably not depend on ptrdiff_t directly, since that is a
POSIX type that could simply be assumed.  However, if an application
has asked for both module A and B, because it want to build on hosts
that doesn't have ptrdiff_t, gnulib-tool must make sure the AH_BOTTOM
for module A is invoked before AH_BOTTOM for module B.

> For example, according to your approach, using the 'gettimeofday' module
> would put a  #include <sys/time.h>  into config.h.
> And using the 'gettime' module put a  #define _GNU_SOURCE 1  into config.h.
> 'gettime' depends on 'gettimeofday' but not vice versa. So by tying the
> the config.h ordering to gnulib module dependencies, the
> #include <sys/time.h>  would come before the  #define _GNU_SOURCE 1.
> Which is wrong.

Right, but I also said #define's (from AC_CHECK_FUNC, AC_DEFINE etc)
should go first.  So gettime would add #define _GNU_SOURCE 1 in the
first part of config.h using AC_GNU_SOURCE, and gettimeofday would add
#include <sys/time.h> using AH_BOTTOM.

But I can see that there may be a potential general problem.  If a
module want to modify the behavior of a header file, that another
module include, the first module must use #define's (and not via
AH_BOTTOM) to do so.  Including a file, via AH_BOTTOM, that contain
the #define that alter the behavior of the other header file, might
end up too late in the process.  I wonder if there are any real
examples of the later, though.

Sorry if I'm unclear, it is sort of difficult to explain this...  and
I may be thinking incorrectly as well.

Thanks.