stdin closed / unavailable during testing

[t-bltg]

#3065 broke Sixel.jl testing (this used to work on 1.6, 1.7, so I consider this a regession).

In our case, we need stdin opened in order to query / analyze terminal properties via control sequences.

Could we set an option to keep stdin opened if we cannot revert the PR ?

cc @johnnychen94, @IanButterworth

[oxinabox]

This also broke DataDeps.jl oxinabox/DataDeps.jl#104
where it is a feature that the user is prompted to confirm that they do infact want to download data during the test.
(which they can bypass by setting an enviroment variable).
I feature I am no longer sure the wisdom of, but given it has been working this way for 5 years, I am not inclined to change.

[oxinabox]

also JuliaCI/PkgTemplates.jl#370

[IanButterworth]

I don't understand what happened here.

#3065 removed stdin forwarding that was added in #2933 (because of #3062). Both of which were during development of 1.8.0

If this used to work before 1.8, the change & revert above isn't the reason for the breakage

[IanButterworth]

1.7.3

Pkg.jl/src/Operations.jl

Line 1630 in 0fae780

p = run(ignorestatus(cmd))

1.8.0

Pkg.jl/src/Operations.jl

Line 1754 in 63f4405

p = run(pipeline(ignorestatus(cmd), stdout = sandbox_ctx.io, stderr = stderr_f()), wait = false)

[IanButterworth]

Ok.

From the Base.run docs

When wait is true (default), I/O streams are shared with the parent process. Use pipeline to control I/O redirection.

@vtjnash in #3062 you said that one shouldn't forward stdin, but isn't that what run was doing by default before 1.8?

[vtjnash]

You should not forward and set wait=false. It is application dependent though if you need stdin

[IanButterworth]

Ok. So would this be better?

p = run(ignorestatus(cmd), stdin, sandbox_ctx.io, stderr_f())

[vtjnash]

The Pkg tests are run on a worker which does not have access to stdin

[vtjnash]

Tests that need a TTY, should be BYO

[IanButterworth]

So is this a way forward?

• revert Remove stdin forwarding from Pkg.test #3065 (re-enable stdin forwarding)
• When Pkg tests actually test Pkg.test we should provide stdin

[IanButterworth]

So is this a way forward?

• revert Remove stdin forwarding from Pkg.test #3065 (re-enable stdin forwarding)
• When Pkg tests actually test Pkg.test we should provide stdin

[IanButterworth]

bump @vtjnash

[vtjnash]

Sure, we could move julia/test/testhelpers/FakePTYs.jl into the test harness itself. You would not want to revert #3065 though, since that brings back all sorts of other problems.