stdout buffer causes messages not to be delivered on arrival

[Justin R]

By default, stdout is buffered. If writing out to a terminal, it will flush after each newline. However, if the stdout is redirected, it won't flush on each newline. It can be flushed manually or the buffer can be disabled.

http://stackoverflow.com/questions/1716296/why-does-printf-not-flush-after-the-call-unless-a-newline-is-in-the-format-strin

Something like this could fix the problem:

int main (int argc, char **argv) {
        setbuf(stdout, NULL);

Or the buffer can be flushed manually after each write.

I made this an issue rather than a PR because I'm not sure of the best way to handle it.

[Magnus Edenhill]

Would a command-line option to turn on unbuffered I/O be sufficient?
I'm not 100% certain unbuffered I/O wont have adverse performance impacts in some cases (e,g. small messages), and would therefor not want to make this default at this point.

[Justin R]

Yea agreed. Making it configurable would enable it to work in either case
👍
On Jun 8, 2014 1:29 PM, "Magnus Edenhill" [e-mail] wrote:

Would a command-line option to turn on unbuffered I/O be sufficient?
I'm not 100% certain unbuffered I/O wont have adverse performance impacts
in some cases (e,g. small messages), and would therefor not want to make
this default at this point.

—
Reply to this email directly or view it on GitHub
#3 (comment).

[Justin R]

Shouldn't there still be a flush after a given timeout though? I can't
think of a case where it'd be OK for data to sit in the stdout buffer
indefinitely unless more data was buffered
On Jun 8, 2014 2:05 PM, "Justin Ratner" [e-mail] wrote:

Yea agreed. Making it configurable would enable it to work in either case
👍
On Jun 8, 2014 1:29 PM, "Magnus Edenhill" [e-mail]
wrote:

Would a command-line option to turn on unbuffered I/O be sufficient?
I'm not 100% certain unbuffered I/O wont have adverse performance impacts
in some cases (e,g. small messages), and would therefor not want to make
this default at this point.

—
Reply to this email directly or view it on GitHub
#3 (comment).

[Magnus Edenhill]

You are probably right, I'll look into it.

[Justin R]

OK thinking about it a little more. A command-line arg to set buffered
true/false would be the simplest solution. Worrying about flushing after a
timeout would probably be getting ahead of ourselves.
On Jun 8, 2014 2:08 PM, "Justin Ratner" [e-mail] wrote:

Shouldn't there still be a flush after a given timeout though? I can't
think of a case where it'd be OK for data to sit in the stdout buffer
indefinitely unless more data was buffered
On Jun 8, 2014 2:05 PM, "Justin Ratner" [e-mail] wrote:

Yea agreed. Making it configurable would enable it to work in either case
👍
On Jun 8, 2014 1:29 PM, "Magnus Edenhill" [e-mail]
wrote:

Would a command-line option to turn on unbuffered I/O be sufficient?
I'm not 100% certain unbuffered I/O wont have adverse performance
impacts in some cases (e,g. small messages), and would therefor not want to
make this default at this point.

—
Reply to this email directly or view it on GitHub
#3 (comment).

[Magnus Edenhill]

Yeah, the downside of -u is the syscall overhead when consuming lots of small messages.
Each message will perform a write() when unbuffered, while in buffered mode a bunch of messages will be accumulated before that write().

[Justin R]

d4d1b89