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

    witty-crayon-22786

    03/17/2020, 5:27 PM
  • h

    hundreds-breakfast-49010

    03/18/2020, 6:26 PM
    so if I run 
    ./v2 --enable-pantsd lint src/python/pants/engine::
    , I get a rust panic error message printed from an 
    unwrap
    deep within the termion code
    h
    w
    • 3
    • 31
  • h

    hundreds-father-404

    03/18/2020, 10:18 PM
    Hi, followup on https://github.com/pantsbuild/pants/pull/9334 that Stu recommended putting in Slack. That PR adds a new 
    targets()
    entry point to 
    register.py
    like this:
    Copy code
    def targets():
    return [PythonBinary, PythonLibrary, PythonTests]
    It leaves open the question of how do you register macros like 
    objects
    and 
    context_aware_object_factories
    . In a discussion with Stu, I was thinking something like this:
    Copy code
    def macros():
  return {"objects": [O1, O2], "context_aware_object_factories": [CAOF1]
    My reasoning is that macros are by definition an “advanced” concept. I would love for plugin authors to be able to register Targets without having to know about macros. Thoughts?
    e
    h
    w
    • 4
    • 45
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:01 PM
    so, is it the case that we always use 
    Nailgun
    -related code whenever we run pantsd, even if we're not doing anything java-related?
  • w

    witty-crayon-22786

    03/18/2020, 11:02 PM
    yes. pantsd is implemented using the nailgun protocol
  • a

    aloof-angle-91616

    03/18/2020, 11:02 PM
    there’s a PR i made a while ago that i think i closed that tried to explicitly separate the two
  • a

    aloof-angle-91616

    03/18/2020, 11:03 PM
    “pailgun”
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:04 PM
    hm, I've seen refrences to pailgun in places in the code
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:05 PM
    why does it make sense to use the Nailgun protocol in non-java-related code?
  • a

    aloof-angle-91616

    03/18/2020, 11:05 PM
    it’s easy to extend
  • w

    witty-crayon-22786

    03/18/2020, 11:06 PM
    it is the most widely used protocol i know of for this usecase: spawning a process and connecting stdio
  • w

    witty-crayon-22786

    03/18/2020, 11:06 PM
    the extensions that we have to it are no longer necessary and can be removed (because we no longer fork)
    • 1
    • 1
  • a

    aloof-angle-91616

    03/18/2020, 11:06 PM
    and has lots of nice reference implementations (jorge wrote one for bloop in scala: “sailgun”)
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:07 PM
    hm, is pantsd running any java code, or is all python implementations of nailgun protocols?
  • w

    witty-crayon-22786

    03/18/2020, 11:08 PM
    pantsd is a python nailgun server
  • w

    witty-crayon-22786

    03/18/2020, 11:08 PM
    and we use a python nailgun client to connect to it.
  • w

    witty-crayon-22786

    03/18/2020, 11:08 PM
    there is a proposal to port pants' client to rust though
    🔥 1
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:09 PM
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:09 PM
    the 
    NailgunClient
    we create there is what connects to the nailgun server?
  • w

    witty-crayon-22786

    03/18/2020, 11:10 PM
    yep
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:48 PM
    one more question - people are talking about how pantsd "no longer forks" - but it looks like the code in 
    process_manager.py
    in 
    daemon_spawn
    is being used right now,and is forking?
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:49 PM
    as in "calls 
    os.fork()
    "?
  • e

    enough-analyst-54434

    03/18/2020, 11:52 PM
    It used to also fork for every request received once it was forked up as a daemon.
  • e

    enough-analyst-54434

    03/18/2020, 11:52 PM
    ... in parallel.
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:53 PM
    ah okay
  • e

    enough-analyst-54434

    03/18/2020, 11:54 PM
    As I recall, all the problems you might expect with mixing threads, signals and forks were encountered.
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:54 PM
    makes sense
  • a

    aloof-angle-91616

    03/18/2020, 11:55 PM
    osx specifically past some 10.x version changed some invariant about forking that caused repeated crashes
  • a

    aloof-angle-91616

    03/18/2020, 11:56 PM
    it makes me sad that forking is no longer a viable strategy in 2020 but honestly it’s to be expected
  • h

    hundreds-breakfast-49010

    03/18/2020, 11:56 PM
    so the code flow here is that we create a 
    NailgunClient
    , which knows how to talk to the forked pantsd process via a port
1...703704705...906Latest