Re: [Patch, fortran] PR29216 & PR29394 - (Redux) Default initialization for derived types that become defined.

[Tobias Burnus <burnus at net-b dot de>]

Hi,

Brooks Moses schrieb:
> THOMAS Paul Richard 169137 wrote:
>> :ADDPATCH fortran:
> Does this actually put things into the automatic patch queue if you
> don't cc to the patches@gcc list?

Yes, and to remove it, you use :REVIEWMAIL:
See http://gcc.gnu.org/wiki/GCC_Patch_Tracking

Actually, :ADDPATCH: also adds a link to this patch into Bugzilla, if a
"PR" is found in the changelog.

> (For that matter, how does one look at the patch queue?)
Click on http://www.dberlin.org/patches
or for Fortran only:
http://www.dberlin.org/patches/patches/list?area=fortran

You can even add patches, which were missed (e.g. no :ADDPATCH:),
manually remove those already checked in or edit those which have ""
rather than "fortran" as area.

> I'd like to put in a small note, for when you start thinking about
> this, that one of the things on my gfortran "wish list" is an option
> to initialize all variables to some exceptional value (e.g., NaN for
> reals) rather than null.  Perhaps that would also be a
> not-too-difficult extension of this?

I actually thought about the same. Having the possibility to initialize
integers as "0", reals as 0.0 and complex as (0.0,0.0) is useful for old
(stupid!) programs, which assume this. But initializing as reals/complex
as NaN is something I miss as it is a nice way to catch uninitialized
variables. (Unfortunally, NaN does not work for integers or logicals.)

(NAG has -nan, ifort has -zero, and g95 has a whole zoo to initialize
all types of variables with an arbitrary value [which I think no one
needs].)

Tobias