fs.stat and fs.statSync may not return identical struct

[DoumanAsh]

• Version: 7.9.0
• Platform: Windows 10 (64 bit)
• Subsystem: fs

Simple code example is here: https://gist.github.com/DoumanAsh/a70b49b3e05aa4c4273dd6917cce13e0
Occasionally found that Stats.dev differs for sync and async versions.
I'm not really sure how it is possible considering that both should use the same OS API...
All other fields are identical

Sync stats output:

Stats {
  dev: 3163614271,
  mode: 16822,
  nlink: 1,
  uid: 0,
  gid: 0,
  rdev: 0,
  blksize: undefined,
  ino: 5066549581028165,
  size: 0,
  blocks: undefined,
  atime: 2017-04-15T06:12:53.786Z,
  mtime: 2017-04-15T06:12:53.786Z,
  ctime: 2017-04-15T06:12:53.786Z,
  birthtime: 2017-04-15T06:03:09.505Z }

Async stat output

Stats {
  dev: -1131353025,
  mode: 16822,
  nlink: 1,
  uid: 0,
  gid: 0,
  rdev: 0,
  blksize: undefined,
  ino: 5066549581028165,
  size: 0,
  blocks: undefined,
  atime: 2017-04-15T06:12:53.786Z,
  mtime: 2017-04-15T06:12:53.786Z,
  ctime: 2017-04-15T06:12:53.786Z,
  birthtime: 2017-04-15T06:03:09.505Z }

[bnoordhuis]

.dev is derived from VolumeSerialNumber, which is a 32 bits unsigned integer. It should fit comfortably in a 64 bits double and it should not (and should not be possible for it to) be less than zero.

Interestingly, 3163614271 and -1131353025 are the same 32 bits bit pattern so this is likely some integer->double conversion bug but it's not immediately obvious to me where it happens. Both the code in libuv and src/node_file.cc look okay at first glance.

[DoumanAsh]

@bnoordhuis i didn't go as far as looking into libuv code, but at least i didn't spot anything in node code itself...

Sadly i don't have possibility to test it on any unix like system, so cannot say if it is windows specific bug or not.

[refack]

Looking into it with windows glasses on

[mscdex]

@DoumanAsh Did you try with previous v7.x versions and/or other major versions?

[DoumanAsh]

@mscdex I remember having this issue in past on some older 7.x version, but cannot say for sure right now.

UPD:
Before v7.7 i get negative integer in both cases
After v7.7 we have current behavior

[seishun]

How did you create the file?

[DoumanAsh]

Just gvim <file>

[mscdex]

The issue is that some optimizations were made to the fs.*statSync() methods awhile back that bring uniformity to how the stat values are represented (they're all doubles now, whereas previously it was all a mixture of 3 different types -- double, signed int, unsigned int). The changes to the fs.*stat() methods were included in a follow-up PR which was marked as semver-major by me because of other unrelated changes in that PR.

So either #11665 has to be backported, or only the fs.*stat()-related changes have to be backported. Also just to note, the positive values should actually be more correct more often than not now (due to larger unsigned value range in double vs signed integer).