[gfortran,patch] Fix intrinsics used as actual arguments

[Paul Thomas]

:REVIEWMAIL:

FX,

Chapeau!

> Hi all,
>
> Here is the patch I promised: this fixes a few problems related to  
> the use of intrinsics as actual arguments. It's a bit long, but it's  
> really straightforward and should be rather easy to review  ;-)

... a "bit" long?!

>
> I'll try to comment my patch below, please ask if you have any  
> question. First, there are the general front-end changes:
>
>   -- the intrinsics list in intrinsic.c was reviewed and the  
> actual_ok field was corrected in each case, according to the  
> Standard. A few GNU intrinsics are kept allowed as actual arguments,  
> for consistency: ACOSH ATANH DACOSH DASINH DATANH DCONJG DIMAG ZABS  
> ZCOS ZEXP ZLOG ZSIN ZSQRT; also, no intrinsic subroutine is allowed  
> as actual argument, so having an actual_ok argument to add_sym_?s  
> functions is useless; the value of actual_ok is propagated into each  
> gfc_intrinsic_sym

I have checked them off against the standard.  All look OK.

>
>   -- create a new function gfc_intrinsic_actual_ok in intrinsic.c  for 
> the resolution functions to perform checks; it is called in  
> resolve_actual_arglist (resolve.c) and an error message is issued if  
> need be; the CHAR intrinsic is special-cased because it's the only  
> one forbidden (as actual arg) in F95 and allowed in F2003

OK

>   -- change gfc_get_extern_function_decl in trans-decl.c to call  
> isym->resolve.f3 when an intrinsic has 3 args

OK

>
>
> Library changes:
>
>   -- make m4 generate new files for AIMAG, ASINH, ACOSH and ATANH  
> specifics
>  -- use a new simple m4 file (misc_specifics.m4) to generate code  for 
> other specifics (including character functions)

The copyright years need changing.  Are these sufficiently different in 
your opinion that you should appear as the contributer?

>
>
> Smaller changes:
>
>   -- a gfc_resolve_index_func function is added to resolve the INDEX  
> intrinsic, so that it can be used as actual argument (PR 16580)

Add a comment as to why you  call it gfc_resolve_index_func, rather than 
gfc_resolve_index, above the entry in intrinsic.c.  I think that it is 
the only exception, is it not?

>   -- intrinsic.texi is updated to mention that DIMAG is a GNU extension
>   -- testcases for all this

OK

>
>
>
> Bootstrapped & regtested on i686-linux, OK for 4.2?

Yes, indeed!  It's more than OK.

Thanks

Paul

PS Exceptionally, I did not apply this to my tree and try it out, since 
it is ready and waiting for the allocatable components patch.