fs: fix realpath{Sync} on resolving pipes/sockets

[ebraminio]

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

Affected core subsystem(s)

fs

Currently

echo "" | node -e "require('fs').realpath('/dev/stdin', console.log)"

and

echo "" | node -e "console.log(require('fs').realpathSync('/dev/stdin'))"

are broken due to our logic of realpath and this patch changes node in a way to match python:

echo "" | python -c "import os; print(os.path.realpath('/dev/stdin'))"

[mscdex]

Use common.isWindows and common.skip() here instead. See here for an example.

[mscdex]

Consider using spawnSync() instead to simplify this a bit and to make future debugging a bit easier.

[ebraminio]

Both done.

[mscdex]

CI: https://ci.nodejs.org/job/node-test-pull-request/8078/

[ebraminio]

This probably needs more study, both of just not following (this patch) and following to the end and raising [current] or not raising error on following symlinks to pipes/sockets should be discussed, lets close it for now.

[ebraminio]

Updated to follow python result.

[ebraminio]

@refack: Hi. Thanks for reviewing my patches. :) Does this need a documentation change also?

[ebraminio]

Have a look at #13130 (comment) also

[refack]

Need documentation.

[refack]

In adition to sanity, I would consider asserting the value of resolvedPath and the returned value from require('fs').realpathSync('/dev/stdin');
Either in the script literal or console.loging it and adding another assert for stdout.

[ebraminio]

Hmm, that resolves to different things on different runs and platforms, that is hard to check I think

[refack]

Than check it has a truthy value if (!resolvedPath) process.exit(2)

[ebraminio]

Done

[refack]

@ebraminio Yes, I would add a note to fs.md#fsrealpathpath-options-callback and realPathSync since this is a deviation from realpath(3).

Something like

*Note*: If `path` resolves to a Socket or a Pipe, the function will return the value of `require('path').resolve(path)`.

P.S. this might be a breaking change (a.k.a semver-major), so I've run CITGM and we'll need to compare
Baseline: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/806/testReport/
With patch: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/802/testReport/
With patch after rebase: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/805/testReport/
With patch after rebase #2: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/808/

[refack]

since this is a deviation from realpath(3).

Or is it converted by the notes?

[ebraminio]

doc is done now, have a look

[refack]

Rebase on master this comment was reverted.

[refack]

BTW is my note actually correct? In that case it's easy to check in the test.

[ebraminio]

No I've copy-pasted your comment just in rush, removed the both as I couldn't reach to anything useful for here.

[refack]

Show me some input -> output results. I'll phrase something better.

[ebraminio]

I am not sure what it should be called, before my patch echo "" | node -e "require('fs').realpath('/dev/stdin', console.log)" was raising error but with my patch now it has a similar result with python, say /proc/12938/fd/pipe:[224961], just on Linux

[refack]

*Note*: If `path` resolves to a socket or a pipe, the function will return a system dependent name for that object.

[ebraminio]

Done

[refack]

CI: https://ci.nodejs.org/job/node-test-commit/10050/

[refack]

So there are a linting error:

not ok 26 - /usr/home/iojs/build/workspace/node-test-linter/test/parallel/test-fs-realpath-pipe.js
message: '''expected'' is assigned a value but never used.'
severity: error
data:
line: 12
column: 7
ruleId: no-unused-vars

and a fail on AIX:

assert.js:92 throw new AssertionError({ ^AssertionError [ERR_ASSERTION]: 1 === 2 at Object.<anonymous> (/home/iojs/build/workspace/node-test-commit-aix/nodes/aix61-ppc64/test/parallel/test-fs-realpath-pipe.js:33:10)

If you don't have a clear idea why this fails on AIX, and since this is not a regression, we could add it to /known_issues/

[ebraminio]

Done. I don't have access to AIX but perhaps it doesn't have the concept of /dev/stdin or that somehow is different there.

[richardlau]

cc @nodejs/platform-aix

[mhdawson]

AIX does not have /dev/stdin. See the second comment in https://unix.stackexchange.com/questions/338667/unix-systems-without-dev-stdin-dev-stdout-and-dev-stderr. I also logged into the community AIX machines and there is no /dev/stdin:

# ls /dev/stdin
ls: 0653-341 The file /dev/stdin does not exist.

So although for redirection behaviour seems to be similar, the files do not actually exist in /dev.

Given that the test depends on /dev/stdin, I'd suggest it just be skipped as it is for windows.

[refack]

AIX does not have /dev/stdin.

@mhdawson Thanks!

Given that the test depends on /dev/stdin, I'd suggest it just be skipped as it is for windows.

I'm +1

[ebraminio]

Given that the test depends on /dev/stdin, I'd suggest it just be skipped as it is for windows.

Done.

[refack]

Pre land CI: https://ci.nodejs.org/job/node-test-commit/10154/

[refack]

arm-fanned fail is infrastructure. Landing.

[refack]

Landed in b3d1e3d

[MylesBorins]

The docs here claim that it is adding a new api... should this have been semver minor?
Backport to v6.x?

[MylesBorins]

ping

[MylesBorins]

ping @refack

[refack]

AFAICT this is semver-patch. It fixes a case that would throw

$ echo""| node -e "console.log(require('fs').realpathSync('/dev/stdin'))"
Error: ENOENT: no such file or directory, lstat '/proc/180/fd/pipe:[140]'
at Object.fs.lstatSync (fs.js:961:11)
at Object.realpathSync (fs.js:1629:21)

so now it returns a sane value:

$ echo""| node -e "console.log(require('fs').realpathSync('/dev/stdin'))"
/proc/577/fd/pipe:[565]

Documentation comment could have been "Pipe/Socket resolve support was added fixed."

[MylesBorins]

@refack this doesn't appear broken on v6.x right now