Feature #9323: IO#writev

• Status: Closed
• Assignee: Glass_saga (Masaki Matsushita)
• Target version: 2.5

[Glass_saga (Masaki Matsushita)]

I propose addition of IO#writev.
It enables gather output from multiple buffers.
If writev(2) is not available, IO#writev uses Array#join to emulate atomic write.

[normalperson (Eric Wong)]

Unless we have use for it, I don't think rb_io_writev should be in
the C API, and even less reason for it to be public for extensions
in ruby/intern.h

Also, it'd probably be good to dedup the fptr->wbuf initialization
code in io_binwrite*.

Otherwise, I think this is fine.

[ioquatix (Samuel Williams)]

I think this is a great idea. It avoids a ton of string issues - e.g. concatenating strings in Ruby before writing them out. It would be great if #write could take an array, and if possible, call writev.

[Glass_saga (Masaki Matsushita)]

• File patch.diff patch.diff added

rebase to trunk

[Glass_saga (Masaki Matsushita)]

• Target version changed from 2.2.0 to 2.5

[Glass_saga (Masaki Matsushita)]

It would be great if #write could take an array, and if possible, call writev.

That's would be nice.

File.open("test", "w") do |f|
  f.write("foo", "bar", "baz") # use writev(2) if possible
end

[akr (Akira Tanaka)]

• Status changed from Open to Feedback

Is there benchmark?

What the situation that writev is actually fast?

[Glass_saga (Masaki Matsushita)]

The main purpose of IO#writev is to make a chance for users to write multiple buffers atomically, not to improve performance.

[normalperson (Eric Wong)]

[e-mail] wrote:

The main purpose of IO#writev is to make a chance for users to
write multiple buffers atomically, not to improve performance.

IO#write on Array#join result is atomic, too; but Array#join can
result in too large buffers and excessive memory use.

I wrote benchmarks for shards io-extra project on Rubyforge
years ago, but I guess that email is no longer available publically.
I've quoted and reposted the test to spew:

https://80x24.org/spew/20170926005509.GA22313@starla/raw

[matz (Yukihiro Matsumoto)]

I vote for making IO#write take multiple arguments, probably using writev(2) inside.

Matz

[Glass_saga (Masaki Matsushita)]

• File patch.diff patch.diff added
• Status changed from Feedback to Assigned
• Assignee set to Glass_saga (Masaki Matsushita)

This patch makes IO#write accept multiple arguments.

[nobu (Nobuyoshi Nakada)]

• Status changed from Assigned to Closed

Applied in changeset trunk|r60351.

---

io.c: fix local variables

• io.c (io_writev): fix local variable declarations, when
  writev(2) is not available. [Feature #9323]

[shyouhei (Shyouhei Urabe)]

• Related to Feature #9420: warn and puts should be atomic added

[shyouhei (Shyouhei Urabe)]

• Related to Feature #14042: IO#puts: use writev if available added