Reporting never ends with mock-fs

[matheuss]

I've seen #246 and #205. Both are closed, but I'm facing this issue right now 😩

Notice the ^ C.

My code: https://github.com/matheuss/hpm/blob/master/test/test.js

[bcoe]

@matheuss the last two times that we fixed this, it related to bundling using npm@v2 vs. npm@v3; since we now have significantly larger dependencies, I'd rather solve the problem for real and take advantage of de-duping.

any help that anyone can provide debugging this gnarly bug, would be greatly appreciated! CC: @novemberborn, @jamestalmage, @addaleax:

1. bundling with npm@v2 (nested dependencies) seems to work with mock-fs.
2. bundling with npm@v3 (top-level dependencies) seems to break.

I think it relates to a shared library, and singleton behavior -- this is a hunch.

[matheuss]

I'm using HyperTerm and it is, sometimes when I run nyc ava, telling me this:

Through iTerm (before clicking ok), I found this:

❯ ~ cd .node-spawn-wrap-98907-3a79ae2824ed
❯ .node-spawn-wrap-98907-3a79ae2824ed ls -la
total 40
drwxr-xr-x    5 matheus  staff   170 Jul 20 10:06 .
drwxr-xr-x+ 112 matheus  staff  3808 Jul 20 10:10 ..
-rwxr-xr-x    1 matheus  staff  4189 Jul 20 10:06 iojs
-rwxr-xr-x    1 matheus  staff  4189 Jul 20 10:06 node
-rw-r--r--    1 matheus  staff   736 Jul 20 10:06 settings.json
❯ .node-spawn-wrap-98907-3a79ae2824ed cat settings.json
{
  "module": "/Users/matheus/.nvm/versions/node/v6.3.0/lib/node_modules/nyc/node_modules/spawn-wrap/index.js",
  "deps": {
    "foregroundChild": "/Users/matheus/.nvm/versions/node/v6.3.0/lib/node_modules/nyc/node_modules/foreground-child/index.js",
    "signalExit": "/Users/matheus/.nvm/versions/node/v6.3.0/lib/node_modules/nyc/node_modules/spawn-wrap/node_modules/signal-exit/index.js"
  },
  "argv": [
    "/Users/matheus/.nvm/versions/node/v6.3.0/lib/node_modules/nyc/bin/wrap.js"
  ],
  "execArgv": [],
  "env": {
    "NYC_CWD": "/Users/matheus/dev/node/hpm",
    "NYC_CACHE": "disable",
    "NYC_SOURCE_MAP": "enable",
    "NYC_INSTRUMENTER": "./lib/instrumenters/istanbul",
    "BABEL_DISABLE_CACHE": 1
  },
  "root": 98907
}

I think that in #205 or #246 I saw something about foreground-child 🤔

[addaleax]

That last one looks like a race condition in the file watcher or in HyperTerm.

[addaleax]

diff --git a/test/test.js b/test/test.js
index 762caa68f1a6..4d5941ff8906 100644
--- a/test/test.js
+++ b/test/test.js
@@ -2,9 +2,12 @@ import {homedir} from 'os';

 import test from 'ava';

-require('mock-fs')({
+const mockfs = require('mock-fs')
+mockfs({
        [`${homedir()}/.hyperterm.js`]: 'module.exports = {plugins: []};'
 });
+test.after(() => mockfs.restore());
+
 require('mock-require')(`${require('os').homedir()}/.hyperterm.js`, './_hyperterm-mocker');

 const api = require('../hyperterm');

should fix it, at least for your test case.

And yeah, sorry, but I don’t think this is something related to nyc’s dependency tree or even just something nyc can do anything about. mock-fs hooks into the fs module from core in a pretty invasive way, that is pretty much impossible to work around from nyc’s side.

There are two solutions for working with mock-fs that I can think of; documenting that calling mockfs.restore() is just plain necessary, or sending the coverage data to the parent nyc process and letting that one write the data to disk.

[matheuss]

Thanks @addaleax! Fixed for me! Never thought of that.

Now with the solution and with what you say, makes sense to state that is not something from nyc. mock-fs indeed change node and we cannot prepare for this.

documenting that calling mockfs.restore() is just plain necessary

This is absolutely necessary. I'll submit a PR.

[jamestalmage]

If someone wanted to take the time to implement tschaub/mock-fs#142, that would solve a lot of headaches as well.