child_process: fork() with shell is impossible?

[vsemozhetbyt]

• Version: from v4. x up to v9.0
• Platform: Windows 7 x64
• Subsystem: child_process

Currently, the doc says nothing if fork() is executed with shell, also no shell option is mentioned. However, fork() is based upon spawn() and almost all the options are transferred as is. So, without shell option we have the default spawn() behavior (without shell):

if (!process.argv[2]) {
  require('child_process').fork(__filename, ['%temp%'], { });
} else {
  console.log(process.argv[2]);
}

%temp%

However, if shell option is set to true, fork() becomes broken in at least two ways:

1. If a path to the executable has spaces, we have this error:

if (!process.argv[2]) {
  require('child_process').fork(__filename, ['%temp%'], { shell: true });
} else {
  console.log(process.argv[2]);
}

>node test.js
'C:\Program' is not recognized as an internal or external command,
operable program or batch file.

2. If a path to the executable has no spaces, we have this error:

>node.8.1.3.exe test.js
child_process.js:106
  p.open(fd);
    ^

Error: EBADF: bad file descriptor, uv_pipe_open
    at Object.exports._forkChild (child_process.js:106:5)
    at Object.setupChannel (internal/process.js:247:8)
    at startup (bootstrap_node.js:53:16)
    at bootstrap_node.js:575:3

So there are some questions:

1. Should we document fork() and shell interaction (and shell option) and fix these issues?
2. If not, should we strip shell option before spawning (and maybe somehow document this)?

[cjihrig]

FWIW, I can't reproduce with master on macOS. Perhaps this is Windows specific.

[bnoordhuis]

If not, should we strip shell option before spawning (and maybe somehow document this)?

Let's do this; fork() + shell doesn't make sense. I've added a good-first-contribution label.