forums.silverfrost.com

mecej4 · Joined: 31 Oct 2006 Posts: 1840

The following trivial cut-down subroutine code causes the compiler to encounter an internal compiler error when the /64 and /checkmate options are used.

PaulLaidler · Posted: Fri Dec 03, 2021 9:45 am Post subject:

mecej4

Thanks for the bug report. It appears that, at the moment /UNDEF works but /FULL_UNDEF (i.e. /CHECKMATE) does not. So the "undef" checking of character variables in this context must be omitted until this is fixed.

PaulLaidler · Posted: Wed Dec 08, 2021 7:02 pm Post subject:

This bug has now been fixed for the next release of FTN95. It concerns setting INTENT(OUT) arrays to the "undefined" state on entry to a subprogram. This is not possible when the array is "star-sized" and the compiler was did not allowing for this case.

At the moment the compiler can only be used with /full_undef (i.e. checkmate) in this context if /inhibit_check 14 is also applied.

JohnCampbell · Joined: 16 Feb 2006 Posts: 2505 Location: Sydney

Paul,

I find it impressive that FTN95 can provide debug information for when the array is "star-sized".
To be able to track the original memory allocation size for this type of declaration is very useful when using SDBG64.

Thanks for maintaining this capability.

mecej4 · Joined: 31 Oct 2006 Posts: 1840

Paul,

Thanks for the bug fix. I am puzzled by your mentioning 'character variables'. My example code did not involve character variables. For that code, if checking cannot be done, a warning message to that effect rather than 'internal compiler error' would suffice.

PaulLaidler · Posted: Thu Dec 09, 2021 7:33 am Post subject:

mecej4

Yes, that was simply my mistake. I had forgotten that /checkmate now includes this INTENT(OUT) checking.