Re: How to get autoconf to respect CC="gcc -std=c89"?

Paul Eggert <[email protected]> writes:

> This doesn't let you say "I want either c89 or c99, but not c11".

That's right, I tried only the very simplest thing. Providing a list
makes sense, thoguh. What should the interface be?

  AC_C_STANDARD_VERSION([c99 c89])

looks reasonable to me; list in order of preference, and whitespace
separation. That's consistent with AC_PROG_CC. But that means that

  AC_C_STANDARD_VERSION([c89 c99])

is a bit meaningless; if compiler has c89, that's what will be used, and
if it doesn't have c89, it won't have c99 either. (Alternative: Let
autoconf automatically sort the list and test for the "largest" language
on the list first. That may be a bit overengineered, though).

I considered trying this out, but then there's some impedance mismatch
between the whitespace separated list to AC_C_STANDARD_VERSION, and the
comma-separated list required by m4_map. Certainly doable, but what's
the easiest way to transform

  [c99 c89]

into 
  
  [99, 89]

?

>> +  m4_ifdef([_AC_C_STANDARD_VERSION_LIST],
>> +    [m4_fatal([AC_PROG_C_STANDARD_VERSION should only be used once], 1)])
>
> Why have this check?

I figured the user should get some error message if using
AC_C_STANDARD_VERSION too late, e.g., after some AC_FOO that implicitly
uses AC_PROG_CC, which will also trigger this check. Not sure of
m4_fatal is the right way to signals error to the user, though. (And
error message could be improved).

> How about something simpler, like replacing this:
>
> m4_map([_AC_PROG_CC_STDC_EDITION_TRY], [[11], [99], [89]])])
>
> with this:
>
> m4_map([_AC_PROG_CC_STDC_EDITION_TRY],
>   [m4_default([_AC_C_STANDARD_VERSION_LIST],
>      [[11], [99], [89]])])

I looked at m4_default, and it looked to me like it checks for empty string, not
undef:ed m4 symbol. If _AC_C_STANDARD_VERSION_LIST is undefined, I
concluded that

  m4_default([_AC_C_STANDARD_VERSION_LIST], whatever)

just expands to the text _AC_C_STANDARD_VERSION_LIST.

So I think we need _AC_C_STANDARD_VERSION_LIST to always be defined. Not
sure where that definition should go, if not in _AC_PROG_CC_STDC_EDITION
itself, but whwreever it is moved, I suspect it would be simpler to jsut
make that definition assign the default value rather than the empty list
needed for m4_default to work as intended.

> and not bothering to check for allowed or duplicated version lists?

Skipping that could make implementation a bit simpler, I guess, at the
cost of less clear user error messages.

Regards,
/Niels

-- 
Niels Möller. PGP key CB4962D070D77D7FCB8BA36271D8F1FF368C6677.
Internet email is subject to wholesale government surveillance.