Re: fortran tests similar with C tests ?

[Steven G. Johnson]

Patrice Dumas wrote:
> > It should not: the architecture is made for these tests to be
> > implementable independently of the language.
>
> Currently, if I'm not wrong AC_CHECK_SIZEOF checks for C, as it
> uses _AC_COMPUTE_INT([(long int) (sizeof ($1))].
> This wont work for fortran and even less in a fortran only environment. 

You're right, Patrice, the test doesn't carry over directly.  I think 
that Akim was hoping that Fortran could use an _AC_COMPUTE_INT-style of 
test.

That would rely, however, on (a) having an expression analogous to 
sizeof(type) in Fortran and (b) implementing AC_LANG_BOOL_COMPILE_TRY in 
Fortran.

(a) I don't know how to do, but maybe there is a way (you could try 
asking on comp.lang.fortran)?

(b) is conceivable, with some re-architecting.  You can maybe use the 
same trick as is used in C, that (all?) Fortran compilers give an error 
if you declare an array with a non-positive size.  e.g. you can 
transform a boolean expression of the form a >= b into a program that 
declares an array of dimension a - b.  I haven't tried this with many 
compilers, though, so I don't know how portable it is...also, it's not 
very useful until you solve (a).

> My trick is rather complicated and fragile, but I had no better idea. It is 
> unsuitable for cross compiling as it uses AC_RUN_IFELSE. It doesn't
> check first that the type is a valid type, but I am not sure that this
> is an issue.

AC_RUN_IFELSE may ultimately be the only way to do things.  (It was how 
the C macro used to check sizes, before people figured out the 
array-declaration trick.)

You assume that your type is a numeric type (you assign it to 1).  It 
will fail, for example, to get sizeof(logical).

Does the Fortran standard require that unformatted writes store the data 
consecutively (i.e. no padding)?

Cordially,
Steven G. Johnson