All versions of `npm init` hang on Node 8.1.0

[iarna]

• Version: 8.1.0
• Platform: Darwin
• Subsystem: Unsure

To reproduce, with Node 8.1.0 and any version of npm (we've explicitly tested w/ 2, 3, 4 & 5):

$ npm init
This utility will walk you through creating a package.json file.
It only covers the most common items, and tries to guess sensible defaults.

See `npm help json` for definitive documentation on these fields
and exactly what they do.

Use `npm install <pkg> --save` afterwards to install a package and
save it as a dependency in the package.json file.

Press ^C at any time to quit.
name: (x) 
version: (1.2.0) 

At the prompt for version Node stops accepting input and doesn't respond to ^Z or ^C. (It does respond to kill -STOP and kill -INT. Continuing the process after a STOP results in keyboard input being echoed but the process still does not run.)

[krokofant]

yarn behaves the same

[addaleax]

This was caused by 81ddeb9, /cc @gibfahn @jasnell

@mscdex I’m removing the v8.x label because this is an issue on master as well.

[Fishrock123]

I can't seem to reproduce on master or 8.1.0?

The WARN does seem to be on the wrong line though?

Press ^C at any time to quit.
package name: (npm-test) npm WARN init canceled
🐟  ~/D/h/npm-test>

[Fishrock123]

I'd be curious to see what e.g. the following patch prints pre/post 81ddeb9

diff --git a/lib/readline.js b/lib/readline.js
index 6a1ed150d7..6113846c91 100644
--- a/lib/readline.js
+++ b/lib/readline.js
@@ -1039,6 +1039,8 @@ function emitKeypressEvents(stream, iface) {
   } else {
     stream.on('newListener', onNewListener);
   }
+  iface.on('close', () => process._rawDebug('## CLOSE'))
+  stream.on('data', () => process._rawDebug('## DATA'))
   if (iface) {
     iface.once('close', () => { stream.removeListener('data', onData); });
   }

[addaleax]

@Fishrock123 I’m not sure I’m reading your output correctly – you first have to enter the package name, only after that I can’t quite anymore.

[iarna]

@addaleax lol! I opened this issue and turned to @zkat and said "I bet Anna sees this and knows exactly the cause"

[Fishrock123]

Oops. 🙊

It appears a close is emitted after pressing enter under normal circumstances...

package name: (npm-test) ## DATA

## CLOSE
version: (0.0.0) ## DATA

description: ## DATA
npm WARN init canceled

[addaleax]

@Fishrock123 Apparently, npm init (or rather the read module) creates a new readline instance per question that it asks…

@jasnell You’re right, reverting is a bit silly, this is easy to fix (in a somewhat obvious way, in hindsight): okay, no, the original patch was incorrect and should be reverted.

--- a/lib/readline.js
+++ b/lib/readline.js
@@ -1040,7 +1040,11 @@ function emitKeypressEvents(stream, iface) {
     stream.on('newListener', onNewListener);
   }
   if (iface) {
-    iface.once('close', () => { stream.removeListener('data', onData); });
+    iface.once('close', () => {
+      stream[KEYPRESS_DECODER] = null;
+      stream[ESCAPE_DECODER] = null;
+      stream.removeListener('data', onData);
+    });
   }
 }

I’ll have a PR and tests up in a bit.

@iarna Naaaaw! 💙

[Fishrock123]

Apparently, npm init (or rather the read module) creates a new readline instance per question that it asks…

While that seems inefficient, readline does apparently emit close at the end of each line: https://nodejs.org/dist/latest-v8.x/docs/api/readline.html#readline_event_close

(Which leads me to believe the patch above may be a bit inefficient too? We can discuss in a PR.)