Zsh Mailing List Archive
Messages sorted by: Reverse Date, Date, Thread, Author

Possible huge setarrvalue optimization



Hello,
in params.c / setarrvalue there is code:

        post_assignment_length = v->start + arrlen(val);
        if (v->end <= pre_assignment_length)
            post_assignment_length += pre_assignment_length - v->end +
            1;

        p = new = (char **) zshcalloc(sizeof(char *)
                                      * (post_assignment_length + 1));

        for (i = 0; i < v->start; i++)
            *p++ = i < pre_assignment_length ? ztrdup(*q++) :
            ztrdup("");
        for (r = val; *r;) {
            /* Give away ownership of the string */
            *p++ = *r++;
        }
        if (v->end < pre_assignment_length)
            for (q = old + v->end; *q;)
                *p++ = ztrdup(*q++);
        *p = NULL;

        v->pm->gsu.a->setfn(v->pm, new);

The point is that the code is also run when doing:

a=( a b c d )
a[1]="x"

The optimization can be: no allocation of new array when size of output
array doesn't change. The problem is that following debug:

        fprintf( _F, "%d %d\n", pre_assignment_length,
        post_assignment_length );

Shows "4 5" for the a[1]="x" assignment. Shows the same for a+=( "x" ).
If this would be fixed, one could detect that pre_assignment_length ==
post_assignment_length, and skip allocation, only do:

        for (r = val; *r;) {
            /* Give away ownership of the string */
            *p++ = *r++;
        }

This would be a huge optimization. Has anyone idea of how the
pre_assignment_length / post_assignment_length computation works, and
why it's wrong for a[1]="x" ?

-- 
  Sebastian Gniazdowski
  psprint@xxxxxxxxxxxx



Messages sorted by: Reverse Date, Date, Thread, Author