On 04/23/2018 08:28 AM, Daniel P. Berrangé wrote: > Rather than specialcasing handling of the '*' character, use fnmatch() > to get normal shell wildcard syntax, as described in 'man glob(7)'. > > To get an indication of the performance impact of using globs instead > of plain string matches, a test program was written. The list of all > 260 log categories was extracted from the source. Then a typical log > filters setup was picked by creating an array of the strings "qemu", > "security", "util", "cgroup", "event", "object". Every filter string > was matched against every log category. Timing information showed that > using strstr() this took 8 microseconds, while fnmatch() took 114 > microseconds. > > IOW, fnmatch is 14 times slower than our existing strstr check. These > numbers show a worst case scenario that wil never be hit, because it s/wil/will > is rare that every log category would have data output. The log category > matches are cached, so each category is only checked once no matter how > many log statements are emitted. IOW despite being slower, this will > be lost in the noise and have no consequence on real world logging > performance. > > Signed-off-by: Daniel P. Berrangé <berrange@xxxxxxxxxx> > --- > src/util/virlog.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > Reviewed-by: John Ferlan <jferlan@xxxxxxxxxx> John BTW: So whether this is available "everywhere" is a bit of an unknown to me - as in strstr would seemingly work on every arch, but fnmatch has this linux-ism wildcard thing going on which leaves a slight bit of doubt in my mind... -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list