Re: [PATCH 1/2] add `#kgio_writev` and `#kgio_trywritev`

[Yura Sokolov <funny.falcon@gmail.com>]

29.05.2012 23:11, Eric Wong написал:
> Sokolov Yura 'funny-falcon<funny.falcon@gmail.com>  wrote:
>> Add methods for using writev(2) syscall for sending array of string in
>> a single syscall. This is more efficient than concatenating strings on
>> Ruby side.
>> `#kgio_trywritev` returns array of strings which are not sent to the
>> socket.
>
> Thanks.  Comments inline.
>
>> Since both methods dups array and strings in it, `#kgio_writev` semantic
>> a bit different from `#kgio_write`: it does not react on changes to
>> array/strings that made in other thread. But I think, this way is more
>> correct.
>
> Perhaps we should use rb_str_locktmp()?

I think, it is good idea :) And use it at `#kgio_write` as well.

I want to redo patch: #kgio_trywritev needs not dup-ing or freeze-ing/lock-ing,
and this method is primary for me.
May be, array should be slice-ed when it is not fully written? It will do
the good job for #kgio_write too.

(I ommit some formatting comments, will fix them)

>
>> +	}
>> +	a->rest_len = a->total_len;
>> +}
>> +
>> +static int writev_check(struct io_args_v *a, long n, const char *msg, int io_wait)
>> +{
>> +	if (a->rest_len == n) {
>> +done:
>> +		a->buf = Qnil;
>> +	} else if (n == -1) {
>> +		if (errno == EINTR) {
>> +			a->fd = my_fileno(a->io);
>> +			return -1;
>> +		}
>> +		if (errno == EAGAIN) {
>> +			if (io_wait) {
>> +				(void)kgio_call_wait_writable(a->io);
>> +				return -1;
>> +			} else if (a->total_len == a->rest_len) {
>> +				a->buf = sym_wait_writable;
>> +			}
>> +			return 0;
>> +		}
>> +		wr_sys_fail(msg);
>> +	} else {
>> +		assert(n>= 0&&  n<  a->rest_len&&  "writev syscall broken?");
>> +		a->rest_len -= n;
>> +		while (n>  0) {
>> +			VALUE str = RARRAY_PTR(a->buf)[0];
>> +			if (RSTRING_LEN(str)>  n) {
>> +				str = rb_str_subseq(str, n, RSTRING_LEN(str) - n);
>> +				RARRAY_PTR(a->buf)[0] = str;
>> +				a->vec->iov_base = RSTRING_PTR(str);
>> +				a->vec->iov_len = RSTRING_LEN(str);
>> +				n = 0;
>> +			} else {
>> +				n -= RSTRING_LEN(str);
>> +				rb_ary_shift(a->buf);
>> +				a->vec++;
>
> Probably better to store the array offset instead of shifting
> Array#shift is O(n).
Well, in Ruby1.9 it is O(1) if not intermixed with other modifying calls.
I'll use rb_ary_subseq, it is also O(1) in Ruby1.9.

>
>> +			}
>> +		}
>> +		return -1;
>> +	}
>> +	return 0;
>> +}
>> +
>> +static VALUE my_writev(VALUE io, VALUE str, int io_wait)
>> +{
>> +	struct io_args_v a;
>> +	long n, iov_cnt, iov_max;
>> +
>> +	prepare_writev(&a, io, str);
>> +	set_nonblocking(a.fd);
>> +	iov_max = sysconf(_SC_IOV_MAX);
>> +retry:
>> +	iov_cnt = RARRAY_LEN(a.buf);
>> +	if (iov_cnt>  iov_max) iov_cnt = iov_max;
>> +	n = (long)writev(a.fd, a.vec, iov_cnt);
>> +	if (writev_check(&a, n, "writev", io_wait) != 0)
>> +		goto retry;
>> +	if (TYPE(a.buf) != T_SYMBOL)
>> +		kgio_autopush_write(io);
>> +	return a.buf;
>
> (I'm not fully awake): You do retry when truncating to iov_max, right?
Retry is done cause writev_check returns -1

>> +static VALUE kgio_writev(VALUE io, VALUE ary)
>> +{
>> +	VALUE array = rb_check_array_type(ary);
>> +#ifdef HAVE_WRITEV
>> +	return my_writev(io, array, 1);
>> +#else
>> +	VALUE str = rb_ary_join(array, Qnil);
>> +	return my_write(io, str, 1);
>> +#endif
>> +}
>
> I don't like #ifdef inside function definitions, so I'd rather have
> something like this:
>
> 	#ifdef HAVE_WRITEV
> 	static VALUE kgio_writev(VALUE io, VALUE ary)
> 	{
> 		...
> 	}
> 	#else
> 	#  include "compat_kgio_writev.h"
> 	#endif
>
Ok

I'll try to redo patch today or tomorrow.

With regards,
Sokolov Yura, aka funny_falcon