https://pantsbuild.org/ logo
Powered by
#general
  • h

    happy-kitchen-89482

    08/18/2015, 5:25 PM
    but as you say, step one is to keep the parsed BUILD files in memory to speed up startup
  • h

    happy-kitchen-89482

    08/18/2015, 5:25 PM
    the rest is in the further future
  • h

    happy-kitchen-89482

    08/18/2015, 5:25 PM
    that first thing is potentially a huge perf win
  • w

    witty-crayon-22786

    08/18/2015, 5:25 PM
    yea
  • w

    witty-crayon-22786

    08/18/2015, 5:26 PM
    kris has his first patches out
  • w

    witty-crayon-22786

    08/18/2015, 5:26 PM
    a lot more is implemented, but he's trying to decide how to partition it
  • w

    witty-crayon-22786

    08/18/2015, 5:26 PM
    and i see him typing so
  • h

    happy-kitchen-89482

    08/18/2015, 5:26 PM
    yeah, I saw the first rb, exciting!
  • r

    rough-minister-58256

    08/18/2015, 5:28 PM
    simple smile - yeah, have filesystem monitoring working end to end via watchman. next is the buildgraph cache
  • b

    bored-art-40741

    08/18/2015, 5:28 PM
    nice
  • h

    happy-kitchen-89482

    08/18/2015, 5:29 PM
    awesome
  • h

    happy-kitchen-89482

    08/18/2015, 5:29 PM
    that plus file fingerprint cache is huge already
  • h

    happy-kitchen-89482

    08/18/2015, 5:29 PM
    even before all the other stuff we could do in a daemon
  • w

    witty-crayon-22786

    08/18/2015, 5:30 PM
    file fingerprint cache?
  • w

    witty-crayon-22786

    08/18/2015, 5:30 PM
    ...oh. as in, watching and fingerprinting all of 
    filedeps
    ?
  • w

    witty-crayon-22786

    08/18/2015, 5:30 PM
    yea, that'd be handy.
  • w

    witty-crayon-22786

    08/18/2015, 5:34 PM
    @rough-minister-58256: i think the locking strategy around the graph is going to be pretty critical... did you decide on a final one?
  • h

    happy-kitchen-89482

    08/18/2015, 5:36 PM
    yeah, not having to re-fingerprint source files all the time just to see if they've changed
  • h

    happy-kitchen-89482

    08/18/2015, 5:36 PM
    since we now know if they've changed thanks to watchman
  • h

    happy-kitchen-89482

    08/18/2015, 5:37 PM
    although we could also solve that by checking the mtime first
  • h

    happy-kitchen-89482

    08/18/2015, 5:38 PM
    I suppose
  • w

    witty-crayon-22786

    08/18/2015, 5:39 PM
    only if we've persisted it somewhere
  • w

    witty-crayon-22786

    08/18/2015, 5:40 PM
    but the stat call isn't free either, so
  • w

    witty-crayon-22786

    08/18/2015, 5:40 PM
    (our internal git uses watchman to make 
    git status
    manageable)
  • w

    witty-crayon-22786

    08/18/2015, 5:40 PM
    (i think it's upstreamed too)
  • r

    rough-minister-58256

    08/18/2015, 5:45 PM
    agreed; haven’t nailed down the locking strategy yet - just getting started on that stuff, have some rough ideas
  • w

    witty-crayon-22786

    08/18/2015, 6:25 PM
    @ancient-stone-50795: when you added the regex for zinc messages, was that because you tried the existing 
    -C-Tnowarnprefixes', '-C${dir}
    thing with zinc and it didn't work?
  • a

    ancient-stone-50795

    08/18/2015, 6:26 PM
    Isn’t 
    -Tnowarnprefixes
    a jmake or twitter compiler wrapper specific thing?
  • a

    ancient-stone-50795

    08/18/2015, 6:27 PM
    I don’t think that argument was recognized by zinc
  • w

    witty-crayon-22786

    08/18/2015, 6:27 PM
    yea, maybe... will check.
1...188189190...2420Latest