cygrunsrv + sshd + rsync = 20 times too slow -- throttled?

[Ken Brown]

On 9/16/2021 11:01 AM, Corinna Vinschen wrote:
> On Sep 16 23:27, Takashi Yano wrote:
>> On Thu, 16 Sep 2021 15:25:46 +0200
>> Corinna Vinschen wrote:
>>> [...]
>>> It would be great if open_setup() would be converted to a method which
>>> is allowed to fail, rather than ignoring errors in sync object creation
>>> and having to test the handle throughout the code.  Given there's only a
>>> single caller of that function (dtable::init_std_file_from_handle), that
>>> shouldn't be much work.  But it's certainly better than ignoring creation
>>> failures in the long run.
>>
>> I do not have confidence to do this correctly in a short time.
>> Leave it as a homework. Or may I leave it to you?
> 
> That's not a short-term problem, but I think it would be good to do this
> with a bit more paranoia at creation time and then rely on the handles
> being available in the subsequent code.  Maybe I can get to it later but
> it certainly wouldn't hurt my feelings if one of you guys look into it
> while I'm abroad. :)
> 
>> I have attached the patches revised.
> 
> LGTM.  I leave it to Ken to take another look and push it.

Done.

One minor thing I just noticed: If I'm not mistaken, query_hdl, hdl_cnt_mtx, and 
reader_closed() are used only for pipes, but they're defined in 
fhandler_pipe_fifo rather than fhandler_pipe.  Is there a reason for that?

Ken