Re: Debug my program please.

Richard E Maine wrote:
Jugoslav Dujic <jdujic@xxxxxxxxx> wrote: 

(snip)

they're pretty safe as long
as you stick to returning a pointer to an already existing target. Actually, I don't see any gotchas with such approach.
If you do, please enlighten me. 

1) I rarely see people doing that. Certainly it is possible and
presumably happens, but I just don't see it much. Pretty much every
sample I see involves allocating new memory, at least as one of the
options. I prefer to simplify my message instead of having to add the
footnote that they are ok as long as you do such functions that way.
Pointers have enough caveats already, and that one gets lost in the
noise, particularly for novice users. Or you write a function with that
rule and then later it gets modified when you (or someone else) is no
longer thinking about that aspect.


Well, one that I know of in some other languages is a hash table
routine.  You pass a pointer to your object, along with the value
to be hashed.  You an later see if something is in the hash table and
it will return a pointer to the object, as passed previously.

It allows one to write the routine independent of the data structure
needed by the user.

2) The observed error rate for such functions is really, really bad.
Seems like almost every time I see a code that uses them, there's a bug
in the usage. To me, this justifies a pretty big hammer for swatting the
bugs. I'd rather overstate the problem if that's what it takes to get
people's attention.


It isn't any better in other languages.  In C if you return a pointer
to newly allocated memory then the caller is expected to free it
later.

3) Not so much a gotcha as a "what is the benefit?" To me, the whole
point of a function is that you can use it as part of an expression.
Life would be far uglier if, for example, instead of
    x = 2.0*sin(t)
you had to write something like
    real :: temp
    ...
    call subroutine_version_of_sin(temp,t)
    x = 2.0*temp
And, of course, it gets far, far worse than this for more complicated
expressions with multiple function references. For a function that
returns a pointer, you are almost always going to reference it like
   p => the_function(whatever)
in which case, you could jut as well have written
   call a_subroutine(p, whatever)
Both do work, but I just nee no advantage to the function form that
outweighs the number of errors I see.

It is a little more convenient in C. With call by value, you have to pass a pointer to a pointer variable as an argument, but a pointer return value is fine.

4) There is a sort of strange footnote that you can add to most of my
personal recommendations, including this one. If someone is enough of an
expert to fully understand why I make the recommendation and to conclude
that the facts of his or her particular situation merit something
contrary to my recommendation, then that is fine. On (rare) occasion I
draw such conclusions myself. I don't think I recall having any code
that violates this one (largely because point 3 above leaves me with
little motivation to), but I've certainly violated some of my other
recommendations when I thought I had enough reason to in a special case.
But.... I figure that if someone needs my advice, then this is the
recommendation they get. Someone who is adequately equipped to
understand and deal with all the caveats shouldn't be needing my advice.
(But then someone so equipped also shouldn't end up needing my help in
debugging any problems that result).


That sounds about right to me.

-- glen

.

Relevant Pages

  • Re: void * vs. int
    ... Either I shift the pointer by PAGE_SHIFT bits, ... Your recommendation between those two would be? ... harmless. ...
    (comp.lang.c)
  • Re: Is anything easier to do in java than in lisp?
    ... Björn Lindberg wrote: ... >> better than no recommendation at all. ... I still think a pointer is better than no pointer, ... It's easy enough to mark it as "doubtful" or whatever. ...
    (comp.lang.lisp)
  • Re: where in registry can i change internet options: general & connect
    ... > where in registry can i change internet options: ... would appreciate advice and pointer. ... and lan settings in connection tab both of which has been ...
    (microsoft.public.windowsxp.network_web)
  • Calling a function by hand
    ... I wanted to get some advice before starting a little project. ... and the pointer to the function are not known until runtime. ... variant open array, like the function Formatdoes. ... I want this to be able to call functions in C, C++, and Delphi. ...
    (borland.public.delphi.language.basm)
  • Re: FAQ?
    ... Section on repairs and maintenance? ... I'm not suggesting we write them out, but a pointer to St. Sheldon? ... adjustment, repair, upgrade and maintenance of all kinds of bicycle. ... However, answers, advice and step by step guides to most common (and ...
    (uk.rec.cycling)