What's the use-case for this?

I embed the postcss-cli watcher in my own application (more info can be found in the webpack-pr I linked). When my application exits it closes stdin for the postcss-cli watcher, however postcss-cli keeps running in the background. On exiting/starting my application multiple times I create multiple orphaned opencss-cli watcher processes. I have to manually kill them.

Why can't your application just send SIGINT/Ctrl+C to the postcss-cli process when it closes?

I think it potentially could, but then it would only be for postcss-cli. An external application that starts postcss-cli usually controls stdin & stdout, therefor it makes most sense to use mechanisms that are already available. Non-daemon applications usually bind to stdin&stdout and therefor it would make sense if it responds to this. The external application is not sending signals, it's just closing and opening stdin/stdout for the child process.
Webpack,Brunch &create-react-app all implement it like this and I'd rather stick to the same solution here.

OK, fair enough. I still think it's a bit of an odd pattern, but if the ecosystem has it, I guess we can do it. This will need an automated test written, though, before I can merge this.

It's quite common for UNIX cli tools to behave like this, shells for example:https://unix.stackexchange.com/questions/110240/why-does-ctrl-d-eof-exit-the-shell

I hope the test is ok like this :-)

It's quite common for UNIX cli tools to behave like this, shells for example:https://unix.stackexchange.com/questions/110240/why-does-ctrl-d-eof-exit-the-shell

This behavior is indeed standard for interactive programs, like shells or REPLs, that are controlled by user input on stdin, I find it odd for a program that does not read stdin for anything else in watch mode, that's all.

Looks great, only one problem; I tested applying the commit with the test to master, and the new test passes even without your original patch, so something with the test isn't working correctly. 🙁

Nice catch! I changed the implementation a bit, the test works properly now. I think thecatch(err) was actually preventing a clean exit which was actually pointing to a flawed implementation from my side. Should be fixed now.

I messed around a bit, the test was sort of failing for the wrong reason earlier; here's a patch to make it better:

diff --git a/test/watch.js b/test/watch.jsindex a6cf2e4..4a51a11 100644--- a/test/watch.js+++ b/test/watch.js@@ -8,6 +8,7 @@ const chokidar = require('chokidar')  const ENV = require('./helpers/env.js') const read = require('./helpers/read.js')+const tmp = require('./helpers/tmp.js')  // XXX: All the tests in this file are skipped on the CI; too flacky there const testCb = process.env.CI ? test.cb.skip : test.cb@@ -287,13 +288,17 @@ testCb("--watch doesn't exit on CssSyntaxError", (t) => { })  testCb('--watch does exit on closing stdin (Ctrl-D/EOF)', (t) => {-  t.plan(0)+  t.plan(1)    const cp = spawn(-    `node ${path.resolve('bin/postcss')} -o output.css -w --no-map`,+    `./bin/postcss test/fixtures/a.css -o ${tmp()} -w --no-map`,     { shell: true }   )+   cp.on('error', t.end)-  cp.on('exit', t.end)+  cp.on('exit', (code) => {+    t.is(code, 0)+    t.end()+  })   cp.stdin.end() })

Copy that text into a file, then rungit apply <file> from the working directory on your branch, commit & push, and this should be good to merge.

Thanks a lot! I'm not familiar enough with the testing framework and checking if thecode = 0 explicitly looks a lot better than just passing it along.
Also,git apply - allows for reading from stdin, no need for the tempfile :-)

Published v8.3.0 🎉

It would have been nice to have this new behaviour only with a new command line argument. This change effectively stopped my docker containers. Luckily this was not a big problem, only surprising. Docker has a simple solution for that:stdin_open: true.

@bug-tape Ah, sorry about that. I didn't think it would affect any existing use-cases, but apparently it did yours. I'll try to be more cautious in the future.