Re: AC_PROG_CC_C99

[Steven G. Johnson]

Paul Eggert wrote:
> Can you please also check for the following features?
>   variable declarations in for loops
>   last member of a struct may be an incomplete array type
>   varargs macros
>   named initialization of structs
>   variable length arrays
>   long long
> These are all features that C99-ish programs are likely to want.

I'm not sure this is a good idea, because:

Paul Eggert also wrote:
> Part of the motivation for keeping that stuff hidden is that we don't
> want people to switch based on whether our macro "thinks" the compiler
> is "C99" or "C89" or "not".  They should switch based on the
> particular feature that they need.

As you yourself point out, the purpose of the AC_PROG_CC_C99 macro is to 
discover compiler flags that put the compiler in a C99-ish mode, *not* 
to guarantee that the compiler completely conforms to the C++ standard.

For example, suppose I want C99 mode because I want to use complex 
arithmetic...and the compiler/libc mostly supports this, but 
variable-length arrays are somewhat broken.  I want AC_PROG_CC_C99 to 
pass and put the compiler in C99 mode, and then I can check for 
complex-arithmetic support.  With your suggestion, AC_PROG_CC_C99 might 
fail because of features I don't need.

AC_PROG_CC_C99 should, ideally, test the minimal set of features that 
pass in C99 mode and fail in non-C99 mode.  In practice, it should test 
slightly more than the minimum in order to have a better chance of 
working on not-yet-tested compilers.  It should certainly be very, very 
far from comprehensive, however.

Steven