unix,stream: fix getting the correct fd for a handle

[saghul]

Checklist

• tests and code linting passes
• a test and/or benchmark is included
• documentation is changed or added
• the commit message follows commit guidelines

Affected core subsystem(s)

stream

Description of change

On OSX it's possible that the fd is replaced, so use the proper libuv
API to get the correct fd.

[kzc]

This is the test case that does not work as expected with node 4.x and 5.x on Mac:

$ node -e 'require("fs").writeSync(process.stdout._handle.fd, "Hello.\n")'

[cjihrig]

Can we translate that into an actual test to prevent regressions?

[saghul]

Also, bonus nachos: we're no longer relying on libuv internals to get the fd.

[saghul]

Can we translate that into an actual test to prevent regressions?

I'm not sure. It would be a OSX only test, which only works if we replaced the fd in the select trick. This could change over time if Apple fixes kqueue to support other fds.

An alternate title would be along the lines of: "use official uv API instead of internal structures".

[kzc]

Because it's a tty test it's more difficult to automate. Spawning, piping and redirecting change the code path within node and libuv pertaining to stdout and stderr.

[cjihrig]

OK, well I did verify that the change worked on my OS X box, so LGTM.

[jasnell]

Nice. LGTM. A test would be good but can definitely understand the constraints.

[bnoordhuis]

I think you might as well remove the #if !defined(_WIN32) guard now.

[saghul]

On Windows that's a HANDLE*, and while it can be cast to an int because only 32 bits are used AFAIS, I wasn't sure if we wanted to expose that.

[bnoordhuis]

Good point. Let's leave it like this, then.

[kzc]

FWIW, dtruss on Mac showed that node 4.x and 5.x process.stdout._handle.fd was returning a closed fd in this case:

node -e 'require("fs").writeSync(process.stdout._handle.fd, "Hello.\n")'

[bnoordhuis]

LGTM

[bnoordhuis]

@saghul I didn't realize it at the time but we have similar logic in src/udp_wrap.cc.

[saghul]

Oh, me neither. I'll make a PR later today if nobody beats me to it.
On May 20, 2016 12:31, "Ben Noordhuis" notifications@github.com wrote:

@saghulhttps://github.com/saghul I didn't realize it at the time but
we have similar logic in src/udp_wrap.cc.

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#6838 (comment)

[Fishrock123]

^ @thealphanerd

[MylesBorins]

@Fishrock123 patch applies cleanly to v4.x-staging will land after it has lived in a release for a week 😄

[MylesBorins]

@saghul is this at all specific to v1.9.0?