Re: PATCH: fortran/26041: FORTRAN compiler won't compile the valid code

[Paul Thomas <paulthomas2 at wanadoo dot fr>]

Hongjiu,

:REVIEWMAIL:

> > Depending on the order of the code, resolve_code may be called before
> > symbols are resolved. This patch tries to resolve symbols before
> > calling resolve_code. Tested on Linux/i686.
>
> It turns out that resolve_code has to be called in the right order. I
> am testing this on Linux/x86-64 now.
This manifestly the right thing to be doing. I have one question and a 
few comments, subject to which this patch is OK to commit.

I have regtested it on FC3/Athlon and have run NIST and all the Meissner 
testcases with it; all looks fine.

The question concerns the recursion into resolve_entries. You do nothing 
that was not done before for this function but since we are about it, I 
wonder if it is meant to visit contained procedures twice or more? Since 
gfc_resolve_type recurses into contained procedures, everybody gets 
visited via the direct call. However, resolve_contained_functions also 
visits contained procedures and calls resolve_entries in the process. 
None of the other function calls can result in such revisits, as far as 
I can tell.

> H.J.
> ----
> 2006-01-31  H.J. Lu  <hongjiu.lu@intel.com>
>
>     PR fortran/26041
>     PR fortran/26064
>     * resolve.c (gfc_resolve_type): New function.
>     (gfc_resolve_code): Likewise.
>     (gfc_resolve): Use them.
Please be a little bit more descriptive; A sentence for each on what the 
new functions do and what was done to gfc_resolve will be enough. eg.:

   * resolve.c (gfc_resolve_type): New function that does everything 
that the
   original gfc_resolve did, except code_resolution.
   (gfc_resolve_code): New function for code resolution that recurses 
through
   contained namespaces.
   (gfc_resolve): Calls gfc_resolve_type and gfc_resolve_code to do 
everything
   previously done by the function of this name.

I am not convinced about the name gfc_resolve_type. At very least, it 
should be gfc_resolve_types. It is a pity that resolve_symbol has 
gone... how about gfc_resolve_entities, to use the terminology of the 
standard? (This is not a drop dead issue for me, though!)

Finally, we have stopped naming testcases pr*****.f90 and give them more 
descriptive names, instead. Look through gfortran.dg to get the idea. 
Also, it would be nice to have a one sentence comment as to what the PR 
was.

Many thanks!

Paul Thomas