Hello!

On Fri, Aug 08, 2014 at 02:56:12AM -0400, crespin wrote:

> Hello,
>
> I don't understand why b->last is used to initialize b->end.
> Why it's not :
> b->end = b->start + size;

There is no real difference, though current code produces 1 less
assembly instruction at least on gcc 4.2 with -O1 / -O2:

.loc 1 39 0
movl %eax, 4(%ebx)
.loc 1 40 0
- addl 12(%ebp), %eax
+ movl 12(%ebp), %eax
+ addl 24(%ebx), %eax
movl %eax, 28(%ebx)
.loc 1 41 0
orb $1, 44(%ebx)

There is no difference in code generated by gcc 4.9, as expected.

Quick look shows that the code dates back to the most recent
snapshot available, and at that time it just used the matching
value which was set last:

h->pre_start = ngx_palloc(pool, size + before + after);
h->start = h->pos.p = h->last.p = h->pre_start + before;
h->end = h->last.p + size;
h->post_end = h->end + after;

Since then the code was simplified, but using of the value that is
set last seems to be still slightly beneficial in some cases.

--
Maxim Dounin
http://nginx.org/

_______________________________________________
nginx mailing list
nginx@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx