Eric Blake wrote: > + strtok_r: use rawmemchr for speed. > + * lib/strtok_r.c (__rawmemchr): Use faster rawmemchr. > + * modules/strtok_r (Depends-on): Add rawmemchr. > ... > Likewise, what about these other modules that only use strchr to find the > terminating byte of a known-terminated string, where rawmemchr could be the > faster choice? > > lib/argz.c: entry = strchr (entry, '\0') + 1; > lib/argz.in.h: __entry = strchr (__entry, '\0') + 1; > lib/canonicalize-lgpl.c: dest = strchr (rpath, '\0'); > lib/canonicalize.c: dest = strchr (rname, '\0'); > lib/glob.c: end_name = strchr (unescape, '\0');
This time I'm with Sam Steingold's argumentation: We should avoid creeping additions of dependencies. rawmemchr is additional code to be compiled on all non-glibc platforms, whereas strchr() is present in libc on all systems. There is nothing wrong with the use of strchr(). I even doubt whether the code in lib/rawmemchr.c can beat assembly language code like the one glibc has for strchr. Furthermore, the code in lib/canonicalize*.c is certainly not speed relevant (since it mostly makes system calls). Bruno
