On Jul 19, 2012 3:38 PM, "Daniel Stone" <<a href="mailto:daniel@fooishbar.org">daniel@fooishbar.org</a>> wrote: > > Hi, > > On 19 July 2012 22:53, Keith Packard <<a href="mailto:keithp@keithp.com">keithp@keithp.com</a>> wrote: > > Adam Jackson <<a href="mailto:ajax@nwnk.net">ajax@nwnk.net</a>> writes: > >> I think there's a combination of factors that pile on to make it worse > >> than you'd expect. > > > > And, caching is completely free, so it seems sensible to just do it > > It's free to not unlink the file, but you still have the overhead of > opening and reparsing the file every time. A more sensible interim > strategy (IMO) would be to always fork xkbcomp once during startup, > preserve the resulting XkbDescRec, then hand that back every time > someone asks for the same keymap, which is pretty much all the time. > It's really not a lot of code either. Don't we already do this? I wrote that patch a couple years ago. You still have the initial xkbcomp fork that's slow on a cold disk, but the in memory keymap is pretty much instant for subsequent devices. I'm on my phone and can't check the archives right now, though. Dan