[Patch, fortran] PR25289 Cannot handle record numbers large than huge(0_4)

[FX Coudert]

Hi Jerry,

I don't have time right now to review that patch. Here are my comments:

> This patch 
> borrows code from trans-types.c to determine whether or not a target 
> supports integer 8 or not.  A new function is added to make this 
> determination.

I think we should use a constant to store that integer kind defined as 
"8 if integer(8) exists and 4 otherwise", to avoid calling a function 
each time we need it. We already have a constant gfc_max_integer_kind in 
place, we could either

  - use that gfc_max_integer_kind, because it is also known from the 
library, as GFC_INTEGER_LARGEST

  - use similar code than the one defining gfc_max_integer_kind to 
define gfc_large_integer_kind as "8 if possible, 4 otherwise"


Index: trans-types.c
===================================================================
--- trans-types.c       (revision 114972)
+++ trans-types.c       (working copy)
@@ -86,6 +86,7 @@

  int gfc_default_integer_kind;
  int gfc_max_integer_kind;
+int gfc_large_integer_kind;
  int gfc_default_real_kind;
  int gfc_default_double_kind;
  int gfc_default_character_kind;
@@ -143,6 +144,13 @@
    /* Set the maximum integer kind.  Used with at least BOZ constants.  */
    gfc_max_integer_kind = gfc_integer_kinds[i_index - 1].kind;

+  if (saw_i8)
+    gfc_large_integer_kind = 8;
+  else if (saw_i4)
+    gfc_large_integer_kind = 4;
+  else
+    gfc_large_integer_kind = gfc_integer_kinds[i_index - 1].kind;
+
    for (r_index = 0, mode = MIN_MODE_FLOAT; mode <= MAX_MODE_FLOAT; mode++)
      {
        const struct real_format *fmt = REAL_MODE_FORMAT (mode);


These are two different choices of our prefered type to transfer 
record-length. My opinion is that we'd better stick with the second 
approach for now.

> On the library side I am sticking with gfc_offset which I think is being 
> handled correctly.  If it is preferred that I do further work on the 
> runtime library side, let me know.

We have this occasion to make the library/front-end "interface" cleaner, 
  so I'd expect we use it fully and use the according type in 
libgfortran. For example, add in libgfortran.h:

#ifdef HAVE_GFC_INTEGER_8
# define GFC_LARGE_INTEGER GFC_INTEGER_8
#else
# ifdef HAVE_GFC_INTEGER_4
#  define GFC_LARGE_INTEGER GFC_INTEGER_4
# else
#  define GFC_LARGE_INTEGER GFC_INTEGER_LARGEST
# endif
#endif

which exactly corresponds to the front-end code (second option) I have 
written above.

About the IOPARM code in itself, I'm not sure how we should handle 
padding at the end of the structure. Have you take this into consideration?

FX