correct pointing in "@projectname --project" testing

[ngam]

• ran into this while customizing conda-forge build of julia
• with @mkitti

Originally, the test as written would always default to home. However, if someone sets a different JULIA_DEPOT_PATH, then the test would fail. This slight edit ensures it passes.

xref: conda-forge/julia-feedstock#157

A simple test to confirm the problem:

JULIA_DEPOT_PATH="/some/path/of/choice:$JULIA_DEPOT_PATH" julia -E 'Base.runtests(["cmdlineargs"]; ncores=ceil(Int, Sys.CPU_THREADS))'

[DilumAluthge]

Won't this error if the user doesn't have JULIA_DEPOT_PATH in their environment?

[ngam]

Won't this error if the user doesn't have JULIA_DEPOT_PATH in their environment?

Ah, you're right. I thought it would populate it automatically. What would you recommend here, then? @DilumAluthge

[DilumAluthge]

1. Can you delete lines 143 and 144 from test/cmdlineargs.jl?
2. Can you squash?

[ngam]

• Can you delete lines 143 and 144 from test/cmdlineargs.jl?
• Can you squash?

I delete the empty lines. Should I still squash now or is it okay to squash upon (potential) merge?

[DilumAluthge]

Let's wait to squash until we're happy with the implementation.

[DilumAluthge]

I would recommend that you do the following:

1. Test my latest suggestion locally to make sure it works for your use case.
2. Accept the suggestion.
3. Rebase and squash.
4. Wait for CI to pass.
5. Ping me once CI is passing, and I can merge.

[DilumAluthge]

Since we are doing env = copy(ENV) and then deleting keys from env, we should use setenv instead of addenv.

Suggested change

	@test expanded == readchomp(addenv(`$exename -e 'println(Base.active_project())'`, "JULIA_PROJECT" => "@foo", "HOME" => homedir(), "JULIA_DEPOT_PATH" => "$(Base.DEPOT_PATH[1])"))
	env = copy(ENV)
	env["JULIA_PROJECT"] = "@foo"
	env["HOME"] = homedir()
	delete!(env, "JULIA_DEPOT_PATH")
	@test expanded == readchomp(setenv(`$exename -e 'println(Base.active_project())'`, env))

[DilumAluthge]

As I mentioned in a different thread:

I think that explicitly deleting the JULIA_DEPOT_PATH environment variable (by doing delete!(env, "JULIA_DEPOT_PATH")) is the most correct solution here, because it makes explicit what we are trying to accomplish.

[mkitti]

Abort this pull request. Do not merge.

@ngam The patch is not needed. As I mentioned in conda-forge/julia-feedstock#157 (comment), I thought this might have already been fixed upstream. It has. It already have been fixed by @DilumAluthge in #42358 .

@KristofferC could we backport #42358 to 1.7 branch?

[mkitti]

The originating error on 1.7.1 is as follows:

Test Failed at $PREFIX/share/julia/test/cmdlineargs.jl:122
  Expression: expanded == readchomp(setenv(`$exename -e 'println(Base.active_project())'`, "JULIA_PROJECT" => "@foo", "HOME" => homedir()))
   Evaluated: "$PREFIX/share/julia/site/environments/foo/Project.toml" == "/home/conda/.julia/environments/foo/Project.toml"

This error occurs when testing 1.7.1 since that still uses setenv.

In #42358 (which is merged on Julia master slated for Julia 1.8), this was changed to addenv, which allows JULIA_DEPOT_PATH to be inherited by the new process. Therefore, we do not need to explicitly set JULIA_DEPOT_PATH due to the use of addenv. Additionally, that PR explicitly tested the use of JULIA_DEPOT_PATH.

The problem that we are trying to address here does not exist on Julia master.

[ngam]

The originating error on 1.7.1 is as follows:

Test Failed at $PREFIX/share/julia/test/cmdlineargs.jl:122
  Expression: expanded == readchomp(setenv(`$exename -e 'println(Base.active_project())'`, "JULIA_PROJECT" => "@foo", "HOME" => homedir()))
   Evaluated: "$PREFIX/share/julia/site/environments/foo/Project.toml" == "/home/conda/.julia/environments/foo/Project.toml"

This error occurs when testing 1.7.1 since that still uses setenv.

In #42358 (which is merged on Julia master slated for Julia 1.8), this was changed to addenv, which allows JULIA_DEPOT_PATH to be inherited by the new process. Therefore, we do not need to explicitly set JULIA_DEPOT_PATH due to the use of addenv. Additionally, that PR explicitly tested the use of JULIA_DEPOT_PATH.

The problem that we are trying to address here does not exist on Julia master.

It seems to me that PR #42358 only addresses this insofar as Base.runtests(["all"]) is called. @DilumAluthge, thoughts?

[DilumAluthge]

Oh I'd totally forgotten about #42358.

Yeah #42358 solved this problem.

[DilumAluthge]

It seems to me that PR #42358 only addresses this insofar as Base.runtests(["all"]) is called.

Hmmm. #42358 should solve this for any invocation of Base.runtests, regardless of the arguments passed to Base.runtests.

[DilumAluthge]

I think the issue here is that #42358 has not yet been backported to 1.6 and 1.7.

[ngam]

It seems to me that PR #42358 only addresses this insofar as Base.runtests(["all"]) is called.

Hmmm. #42358 should solve this for any invocation of Base.runtests, regardless of the arguments passed to Base.runtests.

Sorry, I misread that. Thanks!

[ngam]

Btw @DilumAluthge --- I did my undergrad degree at Brown, in BME. Small world.

[ngam]

Btw @DilumAluthge --- I did my undergrad degree at Brown, in BME. Small world.

@DilumAluthge okay... now... I think I remember you. I believe we took APMA classes together! I graduated in 2015. Anyway, good coincidence, thanks for the great work here and on the other PRs!