Julia v0.7 support

[rdeits]

Includes #139 plus a few more deprecation fixes. Also now checks out WebIO.jl master on 0.7+

Fixes #132
Fixes #145

[piever]

I'm trying to make Interact functional again and am looking for an environment where things are supposed to work (among Blink, Mux, jupyter notebook). Is this PR + Blink the best bet?

[rdeits]

Yup, I have the tests passing locally for Blink with WebIO master and this PR. You'll also need JuliaGizmos/JSExpr.jl#16 on Julia 1.0

[piever]

Cool, on my machine I have the first Interact slider on Julia 1.0! Thank you so much for all the amazing updating work!

[sglyon]

Thank you @rdeits for all the hard work on getting this and related packages working! We owe you

[piever]

Now that WebIO and JSExpr PRs have been merged, I guess this one can also be merged and we can tag Blink 0.8.0 ?

[rdeits]

I'll update this PR to check out WebIO and JSExpr master on travis, just to make sure everything is actually working first.

[rdeits]

Huh, appveyor is consistently failing when trying to uninstall AtomShell:

ERROR: LoadError: IOError: unlink: operation not permitted (EPERM)
Stacktrace:
 [1] uv_error at .\libuv.jl:85 [inlined]
 [2] unlink(::String) at .\file.jl:743
 [3] #rm#9(::Bool, ::Bool, ::Function, ::String) at .\file.jl:253
 [4] #rm at .\none:0 [inlined]
 [5] #rm#9(::Bool, ::Bool, ::Function, ::String) at .\file.jl:263
 [6] #rm at .\none:0 [inlined]
 [7] rm′ at C:\projects\blink-jl\src\AtomShell\install.jl:3 [inlined]
 [8] uninstall() at C:\projects\blink-jl\src\AtomShell\install.jl:12

[rdeits]

Are we perhaps trying to delete a folder instead of a file now? https://stackoverflow.com/a/10979105

[pfitzseb]

We are, but rm′ handles that just fine. I suspect there's still an electron process running when we try to delete the dir. Will look into it.

[pfitzseb]

Ok, so the issue is that Window() creates three electron subprocesses, of which only one is closed by destroy(::Window) (which isn't that surprising). Force closing the others via quit(window.shell) throws an error in JSON.jl, while killing them with kill(window.shell.proc) throws an error somewhere in HTTP.jl

Imho we need a way to gracefully close a Shell and should do that after the tests run, which will fix the test issue on Windows as well. Unfortunately I know too little about error handling in HTTP.jl (or JSON.jl for that matter) to fix this anytime soon.

[piever]

Unrelated: we can stop checking out master of WebIO and JSExpr as they now have releases that support Julia 0.7

[pfitzseb]

Oh, and one slightly annoying thing is that HTTP.jl's info messages are printed when creating a new Window:

julia> w = Window()
Window(1, Electron(Process(`'C:\Users\Basti\.julia\dev\Blink\deps\atom\electron.exe' 'C:\Users\Basti\.julia\dev\Blink\src\AtomShell\main.js' port 5403`, ProcessRunning), Sockets.TCPSocket(Base.Libc.WindowsRawSocket(0x00000000000007f0) active, 0 bytes waiting), Dict{String,Any}("callback"=>##1#2())), Page(3, #undef, Dict{String,Any}("callback"=>##1#2()), Distributed.Future(1, 1, 3, nothing)))

julia> [ Info: Accept:  �    0↑     0↓    0s 127.0.0.1:4714:4714 ≣16
[ Info: Accept:  �    0↑     0↓    0s 127.0.0.1:4714:4714 ≣16
[ Info: Accept:  �    0↑     0↓    0s 127.0.0.1:4714:4714 ≣16
[ Info: Accept:  �    0↑     0↓    0s 127.0.0.1:4714:4714 ≣16
[ Info: Accept:  �    0↑     0↓    0s 127.0.0.1:4714:4714 ≣16
[ Info: Closed:  �    1↑     1↓�   4s 127.0.0.1:4714:4714 ≣16

Is there some way to suppress that?

[rdeits]

Can we wrap the serve() call in Logging.with_logger() and pass the logs to a NullLogger ?

[rdeits]

Another suggestion from Slack: JuliaLang/julia#28229 (comment)

[rdeits]

I implemented the suggestion from JuliaLang/julia#28229 (comment) , and it seems to work fine.

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@3eff499). Click here to learn what that means.
The diff coverage is 90.47%.

@@            Coverage Diff            @@
##             master     #143   +/-   ##
=========================================
  Coverage          ?   85.06%           
=========================================
  Files             ?        9           
  Lines             ?      154           
  Branches          ?        0           
=========================================
  Hits              ?      131           
  Misses            ?       23           
  Partials          ?        0

Impacted Files	Coverage Δ
src/AtomShell/window.jl	81.25% <ø> (ø)
src/content/api.jl	100% <ø> (ø)
src/content/config.jl	100% <100%> (ø)
src/AtomShell/install.jl	100% <100%> (ø)
src/rpc/rpc.jl	78.57% <100%> (ø)
src/content/content.jl	91.66% <100%> (ø)
src/AtomShell/process.jl	80% <75%> (ø)
src/content/server.jl	72.72% <88.88%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3eff499...37a5ca0. Read the comment docs.

[rdeits]

@pfitzseb I don't think the windows uninstall issue is severe enough to block getting an otherwise-working 0.7 version out. Do you agree?

[pfitzseb]

I implemented the suggestion from JuliaLang/julia#28229 (comment) , and it seems to work fine.

That won't work in Juno or IJulia or in any environment that sets up their own logger. I'd be in favour of setting up a BlinkLogger that discards all Info level (and lower) logs from inside HTTP.jl and pipes everything else through to the former current_logger(). Basically something like this.

Actually, HTTP.jl should probably take a logger argument for listen as well which could be used for this. See JuliaWeb/HTTP.jl#189, basically.

@pfitzseb I don't think the windows uninstall issue is severe enough to block getting an otherwise-working 0.7 version out. Do you agree?

Yeah, I agree. There still is the problem of creating zombie processes (at least on Windows, I didn't check on other OSs yet) that can't be closed without getting an error on the Julia side though.

That doesn't have to hold up this PR. so merge at your leisure. Thanks again for the great work :)

[piever]

Merging this and tagging a new release, even if it's not perfect, is very helpful to update the rest of the stack (I need Blink to get Knockout to pass tests on 0.7 and 1.0).

[pfitzseb]

Alright, done.

(Wanted to let @rdeits do the honours of merging, but I guess getting a release out quickly is a good idea)

[NHDaly]

This was a stellar effort! Thank you all! I'm sorry i didn't help more in the process. Great job and thanks! :)