Re: Structure of large link libraries in f95

Charles Russell wrote:
Paul Van Delst wrote:

Charles Russell wrote:

Why would you have to recompile everything? Can't you just recompile the routines you're "monkeying with" and replace them in your library(ies)?


I think you have to recompile a whole module if you change any component
of it. Since my "user interface" is to edit the source, I want fast turnaround.

Yes, but your module only contains interface blocks. (At least, that's what I thought we were talking about here.) It can't

But, if you changed the interface you would have to change the corresponding interface block in your "interface block module(s)".

But why do you want to cast your existing, working libraries, in fortran95 anyway? I use the f77 blas/lapack (and others) libraries in my f95 code and when I use various routines from them, I declare them EXTERNAL and link in the required library -- which was compiled once and pretty much forgotten about


But then you get no type checking. That is OK if you can fall back on ftnchek to find your mistakes, but ftnchek won't work with f95 code. I make lots of mistakes and rely on a highly protected programming environment to find them quickly.

Ah, o.k. The type checking requirement was never stated (at least, I didn't see it, sorry). I never mixed types in procedure calls in f77 back in the day, so type checking of f90/95 has never been that big of a wow-factor for me. The ability to overload procedures via type, however, is the bees knees (IMO). In fact, I've done that with some wrappers to lapack routines (pre-lapack95) when i wanted to have both single and double precision version available with the same interface. Good stuff. Anyway....

Some people, I have been told, get things right the first time.

Yeah, right. (Really? :o)


Old and comfortable do not necessarily translate into good for your feet. :o)


The risk of blisters with new shoes is hard to predict.

True. It takes research and plenty of fittings.

It sounds like you've got everything working the way you want it, with a system set up for quick and easy modification. Why even bother with f95 constructs with your existing library setup? "If it ain't broke...." type of thing.

When the rest of my department switched from fortran to C around 1990, I
figured I'd better prepare a bridge to C. When Lahey later dropped support for its fine f77 compiler, I figured I'd better prepare a bridge to f90. So far I haven't had to cross either bridge, but I still consider it prudent to reconnoitre the route and get some idea of what lies on the other side of the bridge.

Fair enough. Can't argue with being careful.

cheers,

paulv

--
Paul van Delst
CIMSS @ NOAA/NCEP/EMC
.

Relevant Pages

  • Re: Structure of large link libraries in f95
    ... Can't you just recompile the routines you're "monkeying with" and replace them in your library? ... I use the f77 blas/lapack libraries in my f95 code and when I use various routines from them, I declare them EXTERNAL and link in the required library -- which was compiled once and pretty much forgotten about ... When Lahey later dropped support for its fine f77 compiler, I figured I'd better prepare a bridge to f90. ...
    (comp.lang.fortran)
  • Re: Getting the file name from a FILE *
    ... string -- which might perhaps include the filename openned with, ... but as there can be considerable differences in parameter ... passing between non-varargs routines and varargs routines, ... I expect that that recompile / relink step would not be considered ...
    (comp.lang.c)
  • Re: variable length fields for flexibility in subroutines
    ... I also have a linkage ... that describes the interface block there is no need to recompile *all* ... standard COBOL, it has to be a called subroutine, and that brings us to ...
    (comp.lang.cobol)
  • Re: interface blocks
    ... I am considering building a generic interface block with all 3500 subroutines in order to ensure that we are not having argument problems. ... if the routines are seriously ...
    (comp.lang.fortran)