chris@mimsy.UUCP (Chris Torek) (10/29/88)
In article <1730@dataio.Data-IO.COM> bright@Data-IO.COM (Walter Bright) writes:
-I've seen a number of people say that the following:
- p = realloc(p,newsize);
- assert(p); /* fail if out of memory */
-is slower and/or less maintainable than:
- newp = malloc(newsize);
- assert(newp); /* fail if out of memory */
- memcpy(newp,p,(newsize < oldsize ? newsize : oldsize));
- free(p);
- oldsize = newsize;
- p = newp;
Obviously this is false. There is, however, one case where I do use
malloc+copy+free instead of realloc: If the data are important and
must not be destroyed by a failed realloc. The draft standard says
that this is already so; but I have my doubts about current implementations.
--
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163)
Domain: chris@mimsy.umd.edu Path: uunet!mimsy!chrisbright@Data-IO.COM (Walter Bright) (11/01/88)
In article <14232@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes: >There is, however, one case where I do use >malloc+copy+free instead of realloc: If the data are important and >must not be destroyed by a failed realloc. The draft standard says >that this is already so; but I have my doubts about current implementations. In which case, as I said, you write your own realloc(), as in: #if BADREALLOC extern void *mem_realloc(); #else #define mem_realloc(p,oldsize,newsize) realloc((p),(newsize)) #endif and write my_realloc to do the malloc/copy/free. This gives you the advantage of using a well-implemented realloc with a fallback to your own if the system one is bad. I have used a (slightly more complex) variation on this for all my software for years, and have been pleased with the results. I try to code in ANSI as much as possible, and use the macro preprocessor for compatibility with stone age compilers. I also port the ANSI library routines when I need them for some port, instead of trying to do without them.