Allow pass-through of output

[goerz]

Allow to capture stdout without disturbing normal output.

Closes #19

[MichaelHatherly]

Cool, thanks! The name seems fine to me, agreed that "tee" is too jargony.

[goerz]

Yes, agreed :-) How's pass_through, which I just added in the last commit?

[MichaelHatherly]

How's pass_through, which I just added in the last commit?

Probably fine. @mortenpi any other naming suggestions?

[goerz]

@mortenpi This should be ready to merge (if you're okay with pass_through for the name of the keyword argument)

[mortenpi]

This looks great, just a few bikesheds! Do you mind adding a note into CHANGELOG.md as well?

[goerz]

All done!

[mortenpi]

One thing I noticed is that it strips the ASCII colors from the passthrough:

I wonder if some IOContext shenanigans would be sufficient here. But happy to merge this as is though, with this as a "known limitation" though.

[mortenpi]

Oh, it does work if you pass color = true as well:

I guess the question is whether we want the passthrough to be consistent with the "input" or the captured output. Both are pretty reasonable behaviors. If we go with the latter (i.e. current behavior), I only that this interaction with color should be documented.

[goerz]

Good point! It wasn't really something I considered. The current behavior seems ok to me. Indeed, the passthrough is always consistent with the captured output, also in that stdout and stderr are merged. I've added a sentence to clarify that and the color behavior.

[goerz]

Can I get away with not adding an explicit test for the color behavior? ;-)

I've added a test for the color behavior as well

[mortenpi]

Can I get away with not adding an explicit test for the color behavior? ;-)

I've added a test for the color behavior as well

I took the liberty of slightly modifying the tests (adding a stdout/err merging test too, basically, and explicitly checking the color codes).

[mortenpi]

Ugh. I guess the real question is whether we need to support 1.0..

no method matching redirect_stdout(::IOContext{IOStream})

[goerz]

Yeah, I was just fixing that… I think it's just the test that doesn't run on 1.0. The feature itself should be fine.

I'm also okay with testing for the exact escape code, I just wasn't quite sure if they're the same on Windows/Unix, hence just the test for any escape code in the last iteration.

[mortenpi]

Yea, restricting the test to 1.6+ seems fine as well!

[goerz]

I just tested in manually on Julia 1.0. It doesn't crash, but it seems like color=true just has no effect. It's getting a little late for me to hunt that down, though. It seems like such an edge case that I'd be willing to just ignore it.

[mortenpi]

It doesn't crash, but it seems like color=true just has no effect. It's getting a little late for me to hunt that down, though. It seems like such an edge case that I'd be willing to just ignore it.

That is a known, documented limitation 😄

IOCapture.jl/README.md

Lines 64 to 69 in cea2050

	### ANSI color / escape sequences
	On Julia 1.5 and earlier, setting `color` to `true` has no effect, because the [ability to
	set `IOContext` attributes on redirected streams was added in
	1.6](https://github.com/JuliaLang/julia/pull/36688). I.e. on those older Julia versions the
	captured output will generally not contain ANSI color escape sequences.