From: Andy Koppe Date: Thu, 19 Nov 2026 20:58:45 GMT Subject: Re: [PATCH v2 1/4] utf8: add display_mode_esc_sequence_len_counted() Message-ID: In-Reply-To: Dirk Gouders wrote: > > On Thu, Nov 19 2026, Andy Koppe wrote: > > > Dirk Gouders wrote: > >> +/* Same as above, but for a NUL-terminated string "s". */ > >> +size_t display_mode_esc_sequence_len(const char *s) > >> +{ > >> + return display_mode_esc_sequence_len_counted(s, strlen(s)); > >> +} > > > > This is a performance regression compared to the original function. > > Calls that previously returned immediately because *s != '\033', or > > after a few bytes of scanning, now scan the entire string with strlen. > > It turns scanning a whole string for SGRs with repeated calls to the > > function from linear to quadratic. > > > > Therefore I think that if the original implementation isn't going to be > > kept, it would be better not to provide the wrapper, and thereby make > > callers think about what they want to pass as the length. > > > > Uses on strbufs can of course compute it with `sb.buf + sb.len - s`, > > but they don't actually need this safety net, because strbufs are > > guaranteed to be NUL-terminated and the function stops at NULs. > > And the length calculation itself adds a source of errors. > > There is only one user in pretty.c that uses the wrapper. > I can change it to use the _counted version and we could then drop the > wrapper. Junio's v1 already included such a change and I could steal it > from there. > > Would that be OK? No objection from me. I would change the make_line_colors_self_contained() function I'm proposing in [1] to the _counted variant anyway if the original display_mode_esc_sequence_len() is going away, to avoid the performance issue with the wrapper. [1] https://lore.kernel.org/git/e9f88b533dfbd10528c9151e2e87207419d0253d.1794905511.git.andy.koppe@gmail.com