rustc `--color always` doesn't always output color

[JayanAXHF]

Zulip discussion: #t-compiler > Rustc `--color always` doesn't always color output

While working on #150895, I discovered that some parts of rustc that use --color flag don't always output color when the flag is set to always.

Relevant Examples:

rust/compiler/rustc_driver_impl/src/lib.rs

Lines 494 to 498 in 4d38622

	if io::stdout().is_terminal() {
	show_md_content_with_pager(&text, color);
	} else {
	safe_print!("{text}");
	}

rust/src/tools/rustfmt/src/parse/session.rs

Lines 102 to 106 in 4d38622

	let emit_color = if supports_color {
	ColorConfig::from(color)
	} else {
	ColorConfig::Never
	};

The check for terminal support should only occur when the option is set to auto. There may be more examples for this as well. These are just the two that I could find.

[Kivooeo]

@nnethercote, for the first example, this would be nice if you could give some context about this here :3

is this an oversight or intended behavior?

[eggyal]

@Kivooeo: whilst nnethercote was the last to touch that line, it was just moved from elsewhere in the file; joshtriplett introduced stdout().is_terminal() in 34f61dd — but that was just replacing a call to the libc isatty function (via the rustc_log crate and its atty dependency), logic that was first added in 7b0a7fd (#42732) by @cengiz-io in 2017! If anyone can answer whether that was intended to be a special case, I suppose it would be them...?

[eggyal]

Apologies, I suppose the relevant commit/decision is when colour was added to such by @tgross35 in 6a1c10b (#112697).

[Kivooeo]

its fine! thanks for investigation tho

[JayanAXHF]

Looks like in

if io::stdout().is_terminal() {
-    show_content_with_pager(&text);
+    show_md_content_with_pager(&text, color);
} else {
    safe_print!("{text}");
}

the show_content_with_pager was replaced with show_md_content_with_pager(&text, color), so the decision to support color only in terminals could have been unintended.

[tgross35]

Apologies, I suppose the relevant commit/decision is when colour was added to such by @tgross35 in 6a1c10b (#112697).

It's been a long time but I don't think this was intentional. I assume it's a similar case in other places, interaction with --color probably just wasn't considered.

I can't imagine any possible downside to making explain respect --color, so I think you're fine to just do it.