why do addprocs() workers not read my ~/.juliarc.jl

[floswald]

just came across this behaviour, not sure it's intended this way. In the example below you can see that my LOAD_PATH on the master process has 3 entries (i pushed the last one in my ~/.juliarc.jl onto it), whereas on the worker process it has only 2. I assume the worker is not reading ~/.juliarc.jl? any reason for that?

   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "help()" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.3.1-pre+40 (2014-08-28 08:46 UTC)
 _/ |\__'_|_|_|\__'_|  |  release-0.3/795bc31 (fork: 66 commits, 21 days)
|__/                   |  x86_64-apple-darwin13.3.0

julia> addprocs(1)
1-element Array{Any,1}:
 2

julia> @everywhere println(LOAD_PATH)
Union(UTF8String,ASCIIString)["/Users/florianoswald/git/julia/usr/local/share/julia/site/v0.3","/Users/florianoswald/git/julia/usr/share/julia/site/v0.3","/Users/florianoswald/.julia/v0.3"]
    From worker 2:  Union(UTF8String,ASCIIString)["/Users/florianoswald/git/julia/usr/local/share/julia/site/v0.3","/Users/florianoswald/git/julia/usr/share/julia/site/v0.3"]

[ivarne]

Last time .juliarc behaviour was changed was #5076. Seems like this is a case that might have been forgotten then. The question is how .juliarc should work with remote workers.

[elextr]

Do the remote machines necessarily have access to ~/.juliarc?

[ivarne]

Julia could automatically send the ~/.juliarc when she starts remote workers. The remote workers could also load their local ~/.juliarc. The question is what behavior makes sense.

[eschnett]

The workers should read their own .juliarc. If they live on a different file system, then path names may be different for the workers.

[floswald]

you mean you have two separate clusters A and B, one with file system (a), one with file system (b), you launch julia on A and it sets up the cluster across A and B? sounds like science fiction to me, but if that is possible this is definitely worth keeping in mind. is that possible? (i love science fiction.)

[elextr]

The suggestion of @ivarne that the rc file be sent from the head process sounds better, having to manually ensure that the ~/.juliarc is the same on all machines sounds like a recipe for forgetfullness to me.

[JeffBezanson]

Neither one of (1) all workers use node 1's juliarc (2) each worker loads its own local juliarc seems obviously right.

[floswald]

True that. In fact I was just asking if and why it's the case it's not
reading it now. Keeping as is right now is fine by me; but adding a
sentence in the parallel section of the docs would be helpful I think.

On Wednesday, 3 September 2014, Jeff Bezanson notifications@github.com
wrote:

Neither one of (1) all workers use node 1's juliarc (2) each worker loads
its own local juliarc seems obviously right.

—
Reply to this email directly or view it on GitHub
#8196 (comment).