Fortran - Re: gcc 4.0 and Fortran

Author

Re: gcc 4.0 and Fortran

Richard E Maine

2005-05-18, 8:58 pm

In article <slrnd84clk.23s.Ivan.Reid@loki.brunel.ac.uk>,
"Dr Ivan D. Reid" <Ivan.Reid@brunel.ac.uk> wrote:

> I came across this comment in an internal forum today. My
> understanding was that one of g95/gfortran was soon to be incorporated into
> the gcc suite so it doesn't make much sense to me. Can anyone suggest what
> may be being misunderstood here?
>
> "(your gcc 4.0 is a very good example of something that is only usable in
> years - loss of fortran and hence all generators)"

Gee. It is hard to even parse the English, much less the meaning behind
it. The "generators" part is a complete mystery - maybe it made more
sense in context. But to take a *VERY* rough stab in the dark...

The author might have been complaining about g77. It sort of fits in
that g77 was pretty stable (perhaps even too stable in that it was
apparently hard to maintain - I say this based only on recollection of
what others have told me, not on any first-hand data), but gfortran is
still very much in development. While gfortran runs some codes, there
are plenty of others that it doesn't (including many of mine). It isn't
yet suitable as a full replacement for g77.... and the authors very
carefully say exactly that. Maybe in a while, but not yet.

Of course, you *CAN* still get g77. It hasn't disappeared off the face
of the earth. One might even say that the rumors of its death are
greatly exaggerated.

Anyway, that's as close as I can come to guessing what the complaint
might have been about. I'm not going to place any money on my guess
though.

--
Richard Maine | Good judgment comes from experience;
email: my first.last at org.domain | experience comes from bad judgment.
org: nasa, domain: gov | -- Mark Twain

Dr Ivan D. Reid

2005-05-18, 8:58 pm

On Wed, 18 May 2005 11:29:52 -0700, Richard E Maine <nospam@see.signature>
wrote in <nospam-8C6A64.11295218052005@news.supernews.com>:
> In article <slrnd84clk.23s.Ivan.Reid@loki.brunel.ac.uk>,
> "Dr Ivan D. Reid" <Ivan.Reid@brunel.ac.uk> wrote:

[color=darkred]

Sorry about the delay for the OP to appear -- it seems Janet (the
UK academic network) was a bit constipated and it took me a while to
realise that no-one had answered any of my posts since April 28th!
A complaint this morning seems to have things flowing again.
[color=darkred]
> Gee. It is hard to even parse the English, much less the meaning behind
> it. The "generators" part is a complete mystery - maybe it made more
> sense in context. But to take a *VERY* rough stab in the dark...

Sorry about that, too, Richard, I wondered whether or not to expand
or just to let it stand, hoping answers would concentrate on the Fortran
angle, not the applications. The "generators" in this context are
Monte Carlo (I believe!) programmes which generate (representations of)
elementary particles and how they then decay into various possibilities
(decay channels) according to the rules of physics as they are currently
understood. These decay schemes are then fed into simulations of our
detector(s) to produce simulated responses so that various schemes of
reconstructing the "hits" into possible particle tracks can be tested.
(As I showed last year there is a limitation to this; the reconstruction
uses precisely the same "geometry" description as the MC, so errors in
the geometry aren't seen. It was only when I had problems reconstructing
_real_ test-beam particle tracks that several flaws in the geometry
were found.)

> The author might have been complaining about g77. It sort of fits in
> that g77 was pretty stable (perhaps even too stable in that it was
> apparently hard to maintain - I say this based only on recollection of
> what others have told me, not on any first-hand data), but gfortran is
> still very much in development. While gfortran runs some codes, there
> are plenty of others that it doesn't (including many of mine). It isn't
> yet suitable as a full replacement for g77.... and the authors very
> carefully say exactly that. Maybe in a while, but not yet.

Yes, it turns out that was mainly what was meant -- that *g77*
was disappearing, and that its replacement(s) hadn't been verified.

> Of course, you *CAN* still get g77. It hasn't disappeared off the face
> of the earth. One might even say that the rumors of its death are
> greatly exaggerated.

This has been pointed out, as well as the fact that in general
there is an intermediate database (all reconstructions use C++) so using
an older system to generate this is no *great* showstopper. It's also
quite obvious that the generators must be written in g77 (though not
perhaps Fortran 77) and in my naivety I imagine that it's in the bits
added in F90/95 (e.g. array operations) that the newer compilers lack user
confidence, so if these are not _used_ then it makes no odds.

> Anyway, that's as close as I can come to guessing what the complaint
> might have been about. I'm not going to place any money on my guess
> though.

You managed to get it pretty close to the money.

--
Ivan Reid, Electronic & Computer Engineering, ___ CMS Collaboration,
Brunel University. Ivan.Reid@brunel.ac.uk Room 40-1-B12, CERN
KotPT -- "for stupidity above and beyond the call of duty".

Brooks Moses

2005-05-20, 4:00 am

"Dr Ivan D. Reid" wrote:
> On Wed, 18 May 2005 11:29:52 -0700, Richard E Maine <nospam@see.signature>
> wrote in <nospam-8C6A64.11295218052005@news.supernews.com>:
>
> This has been pointed out, as well as the fact that in general
> there is an intermediate database (all reconstructions use C++) so using
> an older system to generate this is no *great* showstopper. It's also
> quite obvious that the generators must be written in g77 (though not
> perhaps Fortran 77) and in my naivety I imagine that it's in the bits
> added in F90/95 (e.g. array operations) that the newer compilers lack user
> confidence, so if these are not _used_ then it makes no odds.

My impression, from reading the gfortran mailing list, is that the areas
of incompleteness are both in the odd corners of the F90/F95 additions
and in the odd corners of the really old extensions. For instance,
there's some weird Hollerith stuff that was recently being discussed,
which apparently a lot of F66-era codes use for string handling. Things
like that may well not be working yet.

- Brooks

--
The "bmoses-nospam" address is valid; no unmunging needed.

Harald Anlauf

2005-05-20, 3:59 pm

Brooks Moses <bmoses-nospam@cits1.stanford.edu> writes:

> My impression, from reading the gfortran mailing list, is that the areas
> of incompleteness are both in the odd corners of the F90/F95 additions
> and in the odd corners of the really old extensions. For instance,
> there's some weird Hollerith stuff that was recently being discussed,
> which apparently a lot of F66-era codes use for string handling. Things
> like that may well not be working yet.

You might want to check the bugzilla entries on ICE's on valid code.
E.g., I would not consider PR 15966 or PR 17917 weird nor belonging to
some odd corner. I have some larger 3rd party code that I would not
even think about touching just to work around such compiler bugs, as
long as I can link mixed gfortran and g77 object files.

Meta-bug report PR 19292 gives an overview of the currently requested
fixes and features lacking gfortran to a replacement for g77, so one
might watch this one before declaring g77 obsoleted.

--
Cheers,
-ha

Richard E Maine

2005-05-20, 3:59 pm

In article <ubacmqezci.fsf@heplix.ikp.physik.tu-darmstadt.de>,
Harald Anlauf <anlauf@hep.tu-darmstadt.de> wrote:

> Brooks Moses <bmoses-nospam@cits1.stanford.edu> writes:
>
[color=darkred]
> You might want to check the bugzilla entries on ICE's on valid code.
> E.g., I would not consider PR 15966 or PR 17917 weird nor belonging to
> some odd corner....

If the problems have only to do with odd corners of f90, then my code
must be full of odd corners. Some might agree, I suppose. :-)

--
Richard Maine | Good judgment comes from experience;
email: my first.last at org.domain | experience comes from bad judgment.
org: nasa, domain: gov | -- Mark Twain

Brooks Moses

2005-05-27, 4:00 am

Richard E Maine wrote:
> In article <ubacmqezci.fsf@heplix.ikp.physik.tu-darmstadt.de>,
> Harald Anlauf <anlauf@hep.tu-darmstadt.de> wrote:
>
>
> If the problems have only to do with odd corners of f90, then my code
> must be full of odd corners. Some might agree, I suppose. :-)

Well, every other corner of the code is odd, no? (The rest being even,
of course.)

Seriously, thanks to you both for correcting my impressions.

- Brooks

--
The "bmoses-nospam" address is valid; no unmunging needed.

Sponsored Links