https://pantsbuild.org/ logo
Powered by
#development
  • r

    rough-minister-58256

    05/11/2016, 9:01 AM
    amazing
  • f

    fancy-queen-20734

    05/11/2016, 4:39 PM
    hm. Maybe we need to make the socket dir path shorter?
  • e

    enough-analyst-54434

    05/11/2016, 4:42 PM
    It needs to not be in the repo tree
  • e

    enough-analyst-54434

    05/11/2016, 4:42 PM
    A user could clone their /repo/to/a/very/deeeeeeeep/path
  • f

    fancy-queen-20734

    05/11/2016, 4:50 PM
    lol
  • w

    witty-crayon-22786

    05/11/2016, 5:10 PM
    @wide-energy-11069: yea, that's exactly the assigned bug
  • w

    wide-energy-11069

    05/11/2016, 5:27 PM
    @enough-analyst-54434: if not in the repo, likely there also needs to be someway for multiple pants instances to avoid stumping on each other.
  • e

    enough-analyst-54434

    05/11/2016, 5:27 PM
    yup
  • e

    enough-analyst-54434

    05/11/2016, 5:27 PM
    hash the repo path to a known length, lots of ways
  • r

    rough-minister-58256

    05/11/2016, 5:30 PM
    hm, good idea - but adds yet another facet of workdir-style management
  • r

    rough-minister-58256

    05/11/2016, 5:31 PM
    would you think this would end up in /tmp or ~ or what?
  • e

    enough-analyst-54434

    05/11/2016, 5:45 PM
    I'd think /var
  • e

    enough-analyst-54434

    05/11/2016, 5:45 PM
    That's standard I think for this sort of thing - pidfiles, socket files
  • e

    enough-analyst-54434

    05/11/2016, 5:47 PM
    On the hierarchy of burdens, 0th would be unix-standard locations, next would be XDG newish standard locations, and probably last would be our custom locations. I'm personally fine with having to maintain these 3 classes, certainly the 1st two seem wise in general from a "standard" tool standpoint
  • r

    rough-minister-58256

    05/11/2016, 5:58 PM
    /var seems to make sense if we were deploying a service, but not so sure that’s a reliable path for userside tooling?
  • r

    rough-minister-58256

    05/11/2016, 5:58 PM
    e.g. were we deploying apache, as root you’d mkdir /var/lib/apache or whatever and use that - not sure as a user there’s a known arbitrarily writable path in /var other than /var/tmp
  • r

    rough-minister-58256

    05/11/2016, 5:59 PM
    iirc, bazel keys all of its on-disk stuff off of $TMPDIR or we
  • e

    enough-analyst-54434

    05/11/2016, 6:03 PM
    No strong opinion from me
  • r

    rough-minister-58256

    05/11/2016, 6:05 PM
    I wonder if it’d make sense to relocate workdirs into such a path wholesale
  • r

    rough-minister-58256

    05/11/2016, 6:06 PM
    so instead of a one-off sub-workdir the whole thing would just end up in ~/.pants or something
  • r

    rough-minister-58256

    05/11/2016, 6:07 PM
    ~/.pants/<buildroot_hash>.workdir/*
  • r

    rough-minister-58256

    05/11/2016, 6:08 PM
    tho history has shown pex’ reliance on 
    ~/.pex
    is problematic for some users that don’t have homedirs etc - tho is configurable
  • e

    enough-analyst-54434

    05/11/2016, 6:12 PM
    I don't see any big issues with that idea, /home or /Users falling back to /tmp if no homedir
  • e

    enough-analyst-54434

    05/11/2016, 6:12 PM
    Although $TMPDIR on OSX is in fact a very deep dir
  • w

    witty-crayon-22786

    05/11/2016, 6:12 PM
    er, by "the workdir", you mean: the entire workdir?
  • r

    rough-minister-58256

    05/11/2016, 6:12 PM
    yeh
  • r

    rough-minister-58256

    05/11/2016, 6:13 PM
    pre-solve the scoot use case
  • w

    witty-crayon-22786

    05/11/2016, 6:13 PM
    we could probably do that for process execution sandboxing/etc in the new engine. but... big change. would probably have two workdirs for a while
  • r

    rough-minister-58256

    05/11/2016, 6:13 PM
    yeah
  • r

    rough-minister-58256

    05/11/2016, 6:14 PM
    maybe we should just make watchman talk over a regular old tcp socket for now
1...909192...906Latest