Interpolate PkgId expression, working around #48

[jmert]

This is the small patch I [semi-]guess-and-checked my way to resolving a similar issue as #48, as mentioned in this comment.

[jmert]

To be specific, my test case looked more like this.

ProjectSpecific.jl:

__precompile__()

module ProjectSpecific

using GenericResearch
import Requires: @require
import Reexport: @reexport

function __init__()
    @require PyPlot="d330b81b-6aea-500a-939a-2ce795aea3ee" begin
        include("plot_map.jl")
        @reexport using .Plotting
    end
end

end

with plot_map.jl having the contents:

module Plotting

using GenericResearch
import PyPlot
import PyCall

const healpy = PyCall.PyNULL()

function __init__()
    copy!(healpy, PyCall.pyimport("healpy"))
end

end

Version info:

julia> versioninfo()
Julia Version 0.7.0-beta.270
Commit 8ab41d6d1f* (2018-07-12 00:57 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-7500U CPU @ 2.70GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.0 (ORCJIT, skylake)

(ProjectSpecific.jl) pkg> status
    Status `Project.toml`
  [438e738f] + PyCall v1.17.1
  [d330b81b] + PyPlot v2.6.0
  [189a3867] + Reexport v0.1.0
  [ae029012] + Requires v0.5.1
    Status `Manifest.toml`
  [9e28174c] + BinDeps v0.8.8
  [3da002f7] + ColorTypes v0.7.2
  [5ae59095] + Colors v0.9.0
  [34da2185] + Compat v0.69.0
  [8f4d0f93] + Conda v0.8.1
  [53c48c17] + FixedPointNumbers v0.5.1
  [682c06a0] + JSON v0.18.0
  [b964fa9f] + LaTeXStrings v1.0.0
  [1914dd2f] + MacroTools v0.4.2
  [30578b45] + URIParser v0.3.1
  [81def892] + VersionParsing v1.1.1
  [2a0f44e3] + Base64
  [ade2ca70] + Dates
  [8bb1440f] + DelimitedFiles
  [8ba89e20] + Distributed
  [b77e0a4c] + InteractiveUtils
  [76f85450] + LibGit2
  [8f399da3] + Libdl
  [37e2e46d] + LinearAlgebra
  [d6f4376e] + Markdown
  [a63ad114] + Mmap
  [44cfe95a] + Pkg
  [de0858da] + Printf
  [3fa0cd96] + REPL
  [9a3f8284] + Random
  [ea8e919c] + SHA
  [9e88b42a] + Serialization
  [1a1011a3] + SharedArrays
  [6462fe0b] + Sockets
  [2f01184e] + SparseArrays
  [8dfed614] + Test
  [cf7118a7] + UUIDs
  [4ec0a83e] + Unicode

[timholy]

Big 👍 for doing this. Bonus points for adding a test that demonstrates this. You could probably crib it from @davidanthoff's test case? You might just need a writepkg2 or something.

"A bug fix is for today. A test is forever." If you add the test then future generations can't do stuff (even accidentally) that breaks it.

[jmert]

Sure, I'll look at turning this in to a test case this afternoon. (Unfortunately, I'm now on campus and have to sadly return to my "MATrix-LABoratory-which-shall-not-be-named"-based research 😭)

[timholy]

Been there myself. Once you start writing code in Julia, it's simply amazing how painful it becomes to write code in most other languages.

[MikeInnes]

It looks like this is probably a subtle base issue around storing the PkgId object directly in a precompiled file (though I'm just guessing). If you can come up with a test case it'd be awesome, but if it's too tricky we can merge without. Would also appreciate a confirmation that this fixes the issue on someone else's setup.

[timholy]

No, it's totally straightforward: it's fixing the same issue I described under "One change you definitely want to keep is hard-coding the PkgId that gets passed to listenpkg" in #46. It just turns out it's more than just listenpkg 😄.

For a test, the bugs that have been reported can easily be turned into test cases. It just comes down to having a module that has another precompiled dependency; that wasn't something I inserted in my new tests.

[MikeInnes]

Ah, I had misunderstood that the first time I read it – but I see now that this fixes the call to require as well. That's pretty weird behaviour but I guess it's related to the serialiser and distributed code.

I'll merge this for now as we need to get the fix out for package authors; we can add a test later. Thanks a lot @timholy and @jmert!