Apply colors only when they are supported

[synalice]

Closes #236.
Closes #237.

This PR wraps all calls to owo_colors::OwoColorize::style with function that calls anstream::AutoStream::choice and checks whenever colors are available for the current stderr.

Conserns

The stderr here is hardcoded it that somewhat bothers me. It is possible for that user of eyre to configure printing all of the error messages into the stderr? Because if eyre ever prints color to the stdout and then the user redirects stdout to the file, this check would detect that stderr is still a terminal and allow ANSI symbols to be printed to the file.

I am also wondering if duplicating style_if_possible in color-spantrace and color-eyre is okay. Should I move it somewhere else?

Tests

I've done some quick testing by running examples and this seems to work well. I don't see any colors in the terminal and there are no special symbols from eyre when redirecting to the file.

Here is a script to run all color-eyre examples and print the output to the file. Mind that RUSTFLAGS=-Awarnings will force everything to rebuild. But that saves you from scrolling past all the deprecation warnings (fixed here btw — #235).

#!/bin/env bash

TEST_OUTPUT_DIR=test-no-color-examples
EXAMPLES='custom_filter custom_section debug_perf github_issue multiple_errors panic_compose panic_hook theme_test_helper theme color-eyre-usage'

rm -r $TEST_OUTPUT_DIR || true
mkdir -p $TEST_OUTPUT_DIR

for example in $EXAMPLES; do
    RUSTFLAGS=-Awarnings cargo run --example $example &>> $TEST_OUTPUT_DIR/$example
done

Might be nice to write some unit tests, but I'd prefer if someone else did it.

About tracing_subscriber

The only ANSI symbols I see are the ones coming from tracing_subscriber (color-eyre/examples/theme_test_helper.rs:35). I did a little digging and found that removing this layer color-eyre/examples/theme_test_helper.rs:49 from registry fixes the issue.

The colors are still there, in the terminal, but not in the file. The only thing that changed visually is that the text is no longer cursive. Although I don't quite understand what this layer does and why is it needed. The documentation is not clear to me.

I think we should just remove this layer from every example to avoid teaching people this flawed behaviour. Let's leave it up to them to enable this layer. This way it won't be eyre's problem, but tracing_subscriber's one.

Rant about owo_colors

To be fair, this all feels like a huge blunder from owo_colors. I feel like it is it who should be responsible for detecting colors. They already have a supports-color feature. But enabling it only gives you a method that you should apply everywhere yourself. I can understand why this feature isn't turned on by default, sure. But once I enable it I kind of expect everything to work out of the box. What if you have thousands of lines of code where colors are added? Do you wrap everything with if_supports_color??? This is nonsense.

[pksunkara]

Ideally, we can just call the style_if_possible function as style. From a developer's perspective, they don't care whether that function is styling or not, depending on the env vars.

[pksunkara]

Pinging @epage so he can get more feedback on how anstream is being used in other crates.

[pksunkara]

Can you please do the following changes:

1. Replace all owo_colors::style() usage with owo_colors::Style::new() (needed for 2nd change)
2. Rename all style_if_possible to style.

[epage]

anstream is a larger dependency for just determining color choice and has specific logic dealing with its auto-adapting to old versions of Windows.

You could pull out the choice logic, calling anstyle_query directly, and adapt to the needs here.

Or eyre could do something like clap and be neutral to this and assume the caller is using anstream with ways of stripping the color.

[pksunkara]

Looks good. I will merge once you can get CI passing.

[pksunkara]

@synalice Can you please fix CI?

[synalice]

Sure.

I kind of forgot about that PR, sowwy.

[SharksT]

Any updates on this ?

[synalice]

I think I fixed all CI issue the last time I touched this MR. Not sure why it says "Some checks haven't completed yet" right now.

[pksunkara]

Weird, can you rebase and push once? If it doesn't work, I will manually merge.

[TheToddLuci0]

@pksunkara Any chance this can be merged?

[pksunkara]

The CI is still failing with multiple different errors.

[pksunkara]

CI still failing. @synalice I will be free this week to review and merge this if you are able to get the CI passing.

[pksunkara]

Still failing. Can you please try all the commands from CI locally first? Using github actions takes a lot of time since a maintainer needs to approve every time.