httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From William A Rowe Jr <wr...@rowe-clan.net>
Subject Re: svn commit: r1715938 - /httpd/httpd/trunk/modules/cache/cache_util.c
Date Tue, 24 Nov 2015 15:49:30 GMT
On Tue, Nov 24, 2015 at 9:28 AM, Eric Covener <covener@gmail.com> wrote:

> On Tue, Nov 24, 2015 at 10:21 AM, Jim Jagielski <jim@jagunet.com> wrote:
> > For these types of paths, where we use a switch/case against
> > the 1st char of the token, the real reason why we used that
> > impl was to avoid the (expected) expensive string comparison.
>
> Maybe naively, I thought we were saving some potential function
> call overhead here rather than some strcasecmp overhead.
>

In part, but also in this specific case;

        while (token) {
***
A switch here short circuits three function calls;
                if (!ap_casecmpstrn(token, "no-cache", 8)) {
                    if (token[8] == '=') {
                        cc->no_cache_header = 1;
                    }
                    else if (!token[8]) {
                        cc->no_cache = 1;
                    }
                }
                else if (!ap_casecmpstr(token, "no-store")) {
                    cc->no_store = 1;
                }
                else if (!ap_casecmpstr(token, "no-transform")) {
                    cc->no_transform = 1;
                }
***
A switch here short circuits four function calls;
                else if (!ap_casecmpstrn(token, "max-age", 7)) {
                    if (token[7] == '='
                            && !apr_strtoff(&offt, token + 8, &endp, 10)
                            && endp > token + 8 && !*endp) {
                        cc->max_age = 1;
                        cc->max_age_value = offt;
                    }
                }
                else if (!ap_casecmpstr(token, "must-revalidate")) {
                    cc->must_revalidate = 1;
                }
                else if (!ap_casecmpstrn(token, "max-stale", 9)) {
                    if (token[9] == '='
                            && !apr_strtoff(&offt, token + 10, &endp, 10)
                            && endp > token + 10 && !*endp) {
                        cc->max_stale = 1;
                        cc->max_stale_value = offt;
                    }
                    else if (!token[9]) {
                        cc->max_stale = 1;
                        cc->max_stale_value = -1;
                    }
                }
                else if (!ap_casecmpstrn(token, "min-fresh", 9)) {
                    if (token[9] == '='
                            && !apr_strtoff(&offt, token + 10, &endp, 10)
                            && endp > token + 10 && !*endp) {
                        cc->min_fresh = 1;
                        cc->min_fresh_value = offt;
                    }
                }
***

Plus the overhead if each function call itself, it is still sensible to
narrow out
several groups of tokens (could even use if s[0]=='n' s[1]=='o' tests)

Does it make sense to expose the ap_tolower_ascii() function?  We could
call it ap_tolower_plaintext() or something else to distinguish that it
isn't
lowercasing extended characters?  Or is the case 'N': case 'n': syntax
easier to follow?

Mime
View raw message