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

    rough-minister-58256

    08/10/2016, 2:12 AM
    guessing Yujie and/or Nick might have the best context on how what to do about this - so will wait for their take. it’s unclear to me if this is actually a bug or legit limitation, but at a minimum “must be a glob” vs defined list literal should get an `engine`+`UX` ticket
  • r

    rough-minister-58256

    08/10/2016, 4:36 AM
    to quantify the impact of ^ to source, this was the scale of change needed to get a clean 
    list ::
    post-RB#4114: 
    47 files changed, 119 insertions(+), 119 deletions(-)
  • r

    rough-minister-58256

    08/10/2016, 4:51 AM
    (in all cases, fileset was a list literal)
  • b

    billowy-beach-68498

    08/10/2016, 4:59 PM
    @rough-minister-58256: although i didn’t make this change i can see the reason behind it. we disallow the use of rel_path if fileset is type FilesetWithSpec because FilesetWithSpec is always matching files/globs relative to BUILD file dir without considering rel_path.
  • r

    rough-minister-58256

    08/10/2016, 6:45 PM
    makes sense - it seems like the constraint here is “ability to be cached in the engine”, but at the expense of effectively killing 
    rel_path
    in bundles
  • r

    rough-minister-58256

    08/10/2016, 6:45 PM
    which is fine imo as long as we’re consciously making that decision vs accidentally
  • r

    rough-minister-58256

    08/10/2016, 6:47 PM
    fwict, `rel_path`+bundle seems a little “dirty” from a purist pov - so perhaps it’s more ideal to kill it in favor of locating referenceable bundles under their literal relative paths
  • f

    fancy-queen-20734

    08/10/2016, 8:13 PM
    @witty-crayon-22786: https://github.com/pantsbuild/pants/issues/3773 for prefork alt roots
  • w

    witty-crayon-22786

    08/10/2016, 8:13 PM
    thanks
  • w

    witty-crayon-22786

    08/10/2016, 8:13 PM
    that one is the most critical at the moment i think
  • w

    witty-crayon-22786

    08/10/2016, 8:54 PM
    @rough-minister-58256: re: `repl`; that's post-fork, right?
  • w

    witty-crayon-22786

    08/10/2016, 8:54 PM
    (just to be clear)
  • r

    rough-minister-58256

    08/10/2016, 8:55 PM
    yeah - the problem is that the nailgun client stuff isn’t currently setup for bidirectional comms, so running 
    ./pants repl <x>
    in the daemon context leaves you with broken input capability to actually use the repl
  • r

    rough-minister-58256

    08/10/2016, 8:55 PM
    it’d just be a blocker to turning the daemon on for users, essentially
  • r

    rough-minister-58256

    08/10/2016, 8:56 PM
    what were you thinking?
  • w

    witty-crayon-22786

    08/10/2016, 8:58 PM
    just wanted to make sure we're definitely accounting for the fact that it will always be a "frontend" task
  • w

    witty-crayon-22786

    08/10/2016, 8:59 PM
    sounds like it, just confirming
  • r

    rough-minister-58256

    08/10/2016, 9:00 PM
    yeh, I think it could make sense eventually to exec frontend tasks directly from the thin client (sans the nailgun bits) at the end of a pantsd-based run
  • r

    rough-minister-58256

    08/10/2016, 9:00 PM
    just aren’t there yet, and this is functionally broken
  • r

    rough-minister-58256

    08/10/2016, 9:11 PM
    unclear if it’d be better to head down that path now - or just patch this up
  • w

    witty-crayon-22786

    08/10/2016, 9:18 PM
    i think you want to be in the forked memory space
  • w

    witty-crayon-22786

    08/10/2016, 9:18 PM
    just not in the daemon process
  • w

    witty-crayon-22786

    08/10/2016, 9:18 PM
    (for ease of consuming the products that were produced)
  • w

    witty-crayon-22786

    08/10/2016, 9:19 PM
    it just definitely won't ever be a "sandboxed" process
  • r

    rough-minister-58256

    08/10/2016, 9:22 PM
    ah, currently the pants run in the daemon context wont’ even start until post-fork so that behavior is inherent - graph construction is the only thing we do synchronously in the daemon today
  • r

    rough-minister-58256

    08/10/2016, 9:23 PM
    and I sort of think “frontend tasks” could just be as simple as an 
    os.exec*
    call against a thing in the dist or perhaps .pants.d dir at the end of the run
  • r

    rough-minister-58256

    08/10/2016, 9:24 PM
    e.g. the pytest runner pex, a repl-entrypoint pex, etc
  • w

    witty-crayon-22786

    08/10/2016, 9:24 PM
    maybe.
  • r

    rough-minister-58256

    08/10/2016, 9:24 PM
    separation of the creation of a thing vs execution
  • r

    rough-minister-58256

    08/10/2016, 9:24 PM
    but might not be as simple as that in the broader case
1...134135136...906Latest