Recursive Runs with Multiple Packages Lose Color

[shellscape]

pnpm version: 4.2.2

Code to reproduce the issue:

Run a recursive command which outputs text with ansi colors stdio. For example, using rollup and ava:

$ pnpm run test --filter ./packages/strip --filter ./packages/replace

Expected behavior:

Expectation is that there exists a means to disable the muting of colors by pnpm and allowing the intended colors and ansi codes to be displayed.

Actual behavior:

The image below demonstrates what actually occurs.

As you can see above, rollups output has been "muted" and greyed. While the image below shows the output from pnpm run build --filter ./packages/strip. Note that colors are correct when only one package is run against.

Additional information:

• node -v prints: 12.3.1
• Windows, OS X, or Linux?: Mac OS

[panoply]

Bumping this.

It would be great if when passing the --color flag via an pnpm command that colours will be inherited/respected, as of now --color will just print white.

I might be wrong, but I do recall reading somewhere that pnpm intentionally mutes/strips stdout when running recursively, don't hold me to that because I may have it confused with another project. Eitherway, I would love to see this.

[zkochan]

We can make the child processes inherit the parent's stdio

pnpm/packages/plugin-commands-script-runners/src/runRecursive.ts

Lines 40 to 43 in 5a51caa

	const stdio = (
	opts.workspaceConcurrency === 1 ||
	packageChunks.length === 1 && packageChunks[0].length === 1
	) ? 'inherit' : 'pipe'

But it will be a mess. Is that ok?

[panoply]

But it will be a mess. Is that ok?

So the pnpm output text when running recursive commands would be sacrificed? ie: we loose path prefix?

[zkochan]

correct

[panoply]

I've only partially explored the code base, at what point does pnpm strip the piped stdio when generating the recursive output? I see that chalk and supports-color is used, can we not leverage the environment variable? In my opinion, the way in which pnpm facilitates logging is a huge part of its aesthetic and the overall pnpm feel (so-to-speak).

[panoply]

For more context, opposed to inheriting, when the --color flag is detected, can we not parse the piped stdio and remove any characters that would cause conflict (eg: (\n\s{0,1}|\-{5,}) while preserving newlines and from here split out and appropriate?

[rhyek]

lerna seems to preserve colors with --parallel and also shows a prefix:

[zkochan]

if Lerna managed to do it, then we can probably as well.

[rhyek]

Any idea when this might get looked at? With microsoft/TypeScript#40584 fixed this is the only issue I'm waiting for to switch over from lerna.

[panoply]

At what point is pnpm stripping the stdio? This one is becoming a little difficult to wrangle lately, which module handles this action?

[ghostdevv]

Does anyone have a solution to this?

[rhyek]

Btw, I was wrong about Lerna handling this correctly. In general, it is my understanding that programs will internally decide wether to output color escape codes and usually they do so by detecting if they are currently attached to a tty. If not, they do not output the codes. Pnpm and lerna will pipe stdout of programs they manage for us so that they can prefix (with color) their output for our benefit. Doing so means the programs will not be running inside a tty directly.

Some programs allow for forcing their output via some environment variable or command line argument. For example, node programs that use https://www.npmjs.com/package/chalk can set FORCE_COLOR=1.

I don't think a general solution applicable to all programs regardless of their support for flags is something pnpm (or lerna) can easily or should do. I might be wrong about this, but I think they would have to run some sort of pseudo terminal (maybe https://github.com/microsoft/node-pty?) per program so they think they're inside a tty, capture their output, and finally output that with the colored prefixes to the real tty.

[zheeeng]

I'm using npm-run-all, it managed output color, can we try its solution?

[woubuc]

So I've tracked down where the colour gets stripped because I ran into this issue and it is a little annoying to lose all colour output.

The short version is that ANSI codes are passed from the script output correctly to pnpm and are maintained all the way through the different logger steps. But right at the end, all ANSI codes are stripped while truncating output lines from lifecycle scripts in packages/default-reporter/src/reporterForClient/reportLifecycleScripts.ts:270. Removing the call to stripAnsi() there fixes the issue (but likely introduces other issues).

This should be replaced with a better solution that will truncate the line while being aware of ANSI escape codes, like for instance the cli-truncate package.

[sebaplaza]

Any advance on this ?

[zkochan]

There was an attempt to make it work but it breaks the output: #4559 (comment)

[DzmitryFil]

+1 very annoying

[pokey]

Some programs allow for forcing their output via some environment variable or command line argument. For example, node programs that use https://www.npmjs.com/package/chalk can set FORCE_COLOR=1.

Is it not risky to run with FORCE_COLOR=1? I believe this environment variable would be inherited by all children of a script. What if a program invokes another program and pipes that output to a file? It would be incorrect / surprising to get color codes there

@rhyek's solution of giving each script its own pty using eg https://github.com/microsoft/node-pty seems like the safest bet. It seems like if pnpm detects that it is attached to a tty, then it should use ptys for each recursive process. But certainly possible I'm missing something here

I'm using npm-run-all, it managed output color, can we try its solution?

I haven't actually tested, but from the docs it looks like color only works when the output isn't prefixed

Maybe the simplest place to start is to have a flag to run that causes it to run scripts sequentially and just output a single line saying which script is running right before running each script? Then each script could just inherit the pty. That would solve the use case where I'm not interested in parallelism so much as an easy way to run a script on all my packages. But maybe such a flag already exists and I missed it?

[valerii15298]

+1 also having problems with colours

[peteromano]

Yea this issue really stagnated..

[FFdhorkin]

There appears to be a regression on this. I have mentioned it in #7228 , but it appears colors are once again being stripped. I was using 8.5.1 and upgraded to 8.9.2 but they're still being removed

[zkochan]

I don't think it is a regression. The suggested solution never worked 100% correctly as I have posted in the comment: #4559 (comment)