[Toybox] [PATCH] grep: add --line-buffered and fix regular buffering.

[Rob Landley]

On 4/11/19 12:18 PM, enh wrote:
> Ping for direction here? I'm happy to implement whichever way you want to go.
> 
> I think we had:
> 
> 1. Don't implicitly flush in all the x* routines, add any missing explicit
> flushes (of which I expect there will be very few). 
>
> 2. Don't use the x* routines as much. 

I recently did:

commit 8f882370be150d80969a1910c20b5d223d084b76
Author: Rob Landley <rob at landley.net>
Date:   Tue Apr 2 15:03:32 2019 -0500

    Have xflush() only flush stdout (that's all it checks errors on),
    and tweak a couple comments.

but removing even that much flushing unless explicitly called is probably fine.

The point of xprintf() and friends is if you have

  yes | head

then "yes" should end when stdout closes. It doesn't have to notice
_imediately_, but generating potentially unbounded data into a pipe shouldn't
keep going an arbitrarily long time, and we disable sigpipe because that's bugs
waiting to happen.

FILE * output flushes itself eventually, and then we'll notice the error. I
think musl's FILE buffer is like 256 bytes and glibc was what, 4k? We don't need
to flush that.

The stdio code flushes when the output includes \n so xputc('\n') is already a
flush inside libc anyway. I should grep or xprintf/xputc/xputs uses and see if
there's an obvious one missing. (The password: prompt is the only one that comes
to mind, toysh will need one too...)

Oh, the other thing I have in this bucket is eliminating get_line(), which can't
be done sanely without input blocking (and is largely why FILE * exists; if you
can't push data back into a filehandle then you either read a byte at a time or
remember the overshoot; get_line() reads a byte at a time and performance sucks
and it makes strace output unreadable).

I initially wrote it because posix-2001 didn't have getline() yet (it was a
glibc extension I didn't want to rely on), but posix-2008 added it, so...

Rob