programming -- keep things simple
in [Fortran]
|
From: Terence on 23 Oct 2006 19:30 I have been reading all the posts for some months and I am very surprised with all the problems planted and the contortions programmers are going through to write their programs. Rule 1. If you have a problem (not a simple question) that you have to expose here you are probably going about your problem-solving algorithm wrongly. And any program is an algorithm in concept, with minor algorithms within the main one. There is USUALLY a simpler way - if you think about it first Rule 2. If you feel you have to call some special service or an external non-fortran routine or program (access to Winteracter or Microsoft's API's for screen control excepted). you are probably doing something wrong. Riule 3. Put yourself in the position of the person later modifying your work because you are no longer around. Would you hate yourself if faced with your coding? Is the algorithm obvious? I have never been faced with a mathematical or language-processing requirement I could not solve with simple programming. The only awkward problems I HAVE met were all related to getting round the 64k byte limit of arrays in early fortran, but even then, there are sparse array methods and list-processing, and the use of external storage offered essentially unlimited arrays in the practical sense. OK, so it was a bit slower, but the results came out earlier too! And on time! If you get the chance, have a look at the old algorithms published and libraries of Fortran code available form the old days, and marvel at the elegance ans simplicity forced on people who had only 64K or memory, or even 4K of memory in thier computers.
From: Elijah Cardon on 24 Oct 2006 03:49 "Terence" <tbwright(a)cantv.net> wrote in message news:1161646242.349155.289220(a)i42g2000cwa.googlegroups.com... >I have been reading all the posts for some months and I am very > surprised with all the problems planted and the contortions programmers > are going through to write their programs. > > Rule 1. If you have a problem (not a simple question) that you have to > expose here you are probably going about your problem-solving algorithm > wrongly. And any program is an algorithm in concept, with minor > algorithms within the main one. There is USUALLY a simpler way - if you > think about it first > > Rule 2. If you feel you have to call some special service or an > external non-fortran routine or program (access to Winteracter or > Microsoft's API's for screen control excepted). you are probably doing > something wrong. > > Riule 3. Put yourself in the position of the person later modifying > your work because you are no longer around. Would you hate yourself if > faced with your coding? Is the algorithm obvious? > > I have never been faced with a mathematical or language-processing > requirement I could not solve with simple programming. The only awkward > problems I HAVE met were all related to getting round the 64k byte > limit of arrays in early fortran, but even then, there are sparse array > methods and list-processing, and the use of external storage offered > essentially unlimited arrays in the practical sense. OK, so it was a > bit slower, but the results came out earlier too! And on time! > > If you get the chance, have a look at the old algorithms published and > libraries of Fortran code available form the old days, and marvel at > the elegance ans simplicity forced on people who had only 64K or > memory, or even 4K of memory in thier computers. I've always been looking for good advice. Everything I read here seemed wrong to me. As if it were an exercise in english .... EC
From: Dylan Sung on 24 Oct 2006 05:22 "Terence" <tbwright(a)cantv.net> wrote in message news:1161646242.349155.289220(a)i42g2000cwa.googlegroups.com... > If you get the chance, have a look at the old algorithms published and > libraries of Fortran code available form the old days, and marvel at > the elegance ans simplicity forced on people who had only 64K or > memory, or even 4K of memory in thier computers. > When tackling a problem, I find it easier to create separate programs to do single jobs that spits out the results. Usually, they are only about a hundred or so lines long, so it isn't too hard to figure out what its all about despite the lack of comments, as I do space out block of related stuff, but don't indent. I find indented code harder to read. Dyl.
From: beliavsky on 25 Oct 2006 12:47 Terence wrote: > I have been reading all the posts for some months and I am very > surprised with all the problems planted and the contortions programmers > are going through to write their programs. > > Rule 1. If you have a problem (not a simple question) that you have to > expose here you are probably going about your problem-solving algorithm > wrongly. And any program is an algorithm in concept, with minor > algorithms within the main one. There is USUALLY a simpler way - if you > think about it first > > Rule 2. If you feel you have to call some special service or an > external non-fortran routine or program (access to Winteracter or > Microsoft's API's for screen control excepted). you are probably doing > something wrong. I disagree. I'd guess that many, if not most, "production" Fortran programs need to do this, which is one reason why Fortran 2003 standardizes interoperability with C and why Fortran 2008 has EXECUTE_COMMAND_LINE. That "math" code should be separated from the code using these features. <snip> > If you get the chance, have a look at the old algorithms published and > libraries of Fortran code available form the old days, and marvel at > the elegance ans simplicity forced on people who had only 64K or > memory, or even 4K of memory in thier computers. They may have used EQUIVALENCE, which is not elegant or simple to follow.
From: Ian Bush on 26 Oct 2006 05:58
beliavsky(a)aol.com wrote: > > Terence wrote: >> Rule 2. If you feel you have to call some special service or an >> external non-fortran routine or program (access to Winteracter or >> Microsoft's API's for screen control excepted). you are probably doing >> something wrong. > > I disagree. I'd guess that many, if not most, "production" Fortran > programs need to do this, which is one reason why Fortran 2003 > standardizes interoperability with C and why Fortran 2008 has > EXECUTE_COMMAND_LINE. That "math" code should be separated from the > code using these features. Hmmmm, I think I disagree with both ! For me it would be Rule 2. If you feel you have to call some special service or an external non-fortran routine or program (MPI excepted). you are probably doing something wrong. I've contributed heavily to three "production" Fortran programs which are sold, provided a fair bit of input for a fourth, and bits and pieces for others, and only one of these breaks the above - and that is really a function of the large amount of legacy code in it, Ian |