Using node with process.stdin inside spawned shell

[qfox]

• Version: v8.4.0
• Platform: Darwin duplo 15.6.0 Darwin Kernel Version 15.6.0: Tue Apr 11 16:00:51 PDT 2017; root:xnu-3248.60.11.5.3~1/RELEASE_X86_64 x86_64
• Subsystem: net, stream or child_process (?)

Tested with:

• GNU bash, version 3.2.57(1)-release (x86_64-apple-darwin15)
• GNU bash, version 4.2.25(1)-release (x86_64-pc-linux-gnu)

Not sure is this the same or another bug but in my environment it throws:

Error: write EPIPE
    at _errnoException (util.js:1041:11)
    at WriteWrap.afterWrite [as oncomplete] (net.js:858:14)

But when we played with it a little:

Error: This socket has been ended by the other party
    at Socket.writeAfterFIN [as write] (net.js:352:12)
    at __dirname (/Users/qfox/repos/trendbox-ci/omg.js:11:11)
    at <anonymous>

How to repro:

const { spawn } = require('child_process');
const pause = t => new Promise(resolve => setTimeout(resolve, t));

const { stdin } = spawn('bash', [/*'-i'*/], { stdio: ['pipe', 'inherit'] });

(async () => {
    stdin.write('echo All is fine here\n');
    await pause(200);
    stdin.write('node -e process.stdin && echo After process.stdin\n');
    await pause(2000); // It's important: the bigger pause, the more frequently occurs
    stdin.write('echo After all\n');
})().catch(e => console.error(e.stack));

If we remove usage of process.stdin or force shell to be interactive (pass in ['-i'] flag) or just call node -e process.stdin with </dev/stdin — then the problem disappears.

Would be nice to have a good error message if it's unfixable ;'-(

upd:
See also: #947, #2339, #13278

[gireeshpunathil]

consistently recreatable in Linux, with this code:

var child = require('child_process').spawn('bash')
child.stdin.write('node -e process.stdin\n')
setTimeout(() => {
  child.stdin.write('echo hello world!\n')
},1000)

strace shows that bash exited with
read(0, 0xecbf60, 1) = -1 EAGAIN (Resource temporarily unavailable)

while trying to read the second command, after the sleep.

EAGAIN for read manual says:
The file descriptor fd refers to a socket and has been marked non-blocking (O_NONBLOCK), and the read would block.

So looks like bash is handing over its fd 0 (a socket in this case) it got from node as it is to the child node.

[pid 5529] fstat(0, {st_mode=S_IFSOCK|0777, st_size=0, ...}) = 0

so Node makes it non-blocking to fit its use case

[pid 5529] ioctl(0, FIONBIO, [1]) = 0

back in bash, it waits to read from fd 0, but the read returns with soft error (EAGAIN). This means no data now, please wait.

Unfortunately bash seems to be not prepared for this scenario. So it exits with error.

With this understanding, it is easy to recreate without Node in the picture:

#cat 18446.cc

#include <sys/ioctl.h>
int main() {
  int set = 1;
  ioctl(0, FIONBIO, &set);
  return 0;
}

#g++ 18446.cc
#bash -c './a.out && read'
bash: line 0: read: read error: 0: Resource temporarily unavailable

Should we restore the standard streams on exit? I thought I have seen a PR on this line.

/cc @bnoordhuis

[gireeshpunathil]

linking with #14752 and #17737

[gireeshpunathil]

This issue has been resolved via #20592 and I have verified with both the test programs.

#node 18446.js

All is fine here
After process.stdin
After all

Closing this, fix will mostly be available in the next current release.

[qfox]

Thank you very much!

[gireeshpunathil]

unfortunately #20592 was reverted in master, so re-opening this.

[bnoordhuis]

This should have been fixed by #24260 so I'll go ahead and close it out.