Re: 3.3.0: Possible regression in cygwin DLL (Win10); fixed in snapshot

[Corinna Vinschen <corinna-cygwin@cygwin.com>]

On Nov 11 22:20, Takashi Yano wrote:
> On Thu, 11 Nov 2021 21:02:34 +0900
> Takashi Yano wrote:
> > On Thu, 11 Nov 2021 12:33:21 +0100
> > Corinna Vinschen wrote:
> > > On Nov 11 20:12, Takashi Yano wrote:
> > > > > [FileProcessIdsUsingFileInformation]
> > > > 
> > > > Thanks for advice. get_query_hdl_per_* is called in the write side,
> > > > so only knows pipe handle of the write side. We would like to know
> > > > ProcessId which have pipe handle of the read side. Can we use
> > > > FileProcessIdsUsingFileInformation for this perpose?
> > > 
> > > I don't know.  I just stumbled over it yesterday and I thought it might
> > > be something we could utilize.  Supposedly it returns PIDs for processes
> > > having that file open, but how this works for pipe read/write sides
> > > needs testing.  Of course, knowing how Windows functions usually only go
> > > half the way, the function will either only return the current side of
> > > the pipe, or even an error code :-/  Never mind, it was just an idea.
> > 
> > I have tested the behaviour of FileProcessIdsUsingFileInformation
> > just now. We can use it! I will try to use it in get_query_hdl().
> 
> I have tried to utilize FileProcessIdsUsingFileInformation in
> get_query_hdl_per_process() as the patch attached. It works as
> expected.
> 
> I also measured the response of select() using these functions.
> The first time and second time responses are measured. The second
> time should be much faster than the first time because search
> result has been cached.
> 
> First time, Second time
>  4.620400 [msec], 0.102300 [msec] << get_query_hdl_per_process()
> 19.080400 [msec], 0.199000 [msec] << get_query_hdl_per_system()
> 14.364300 [msec], 0.156800 [msec] << FileProcessIdsUsingFileInformation
> 
> Unfortunately, FileProcessIdsUsingFileInformation is slower than
> current get_query_hdl_per_process(). It takes about 3 times longer
> time.

*laughing manically*

Your previous mail was a nice surprise, but now we're back on earth it
seems.

On second thought I guess this even makes sense.  At Vista times
NtQueryInformationProcess(ProcessHandleInformation) didn't exist,
so the method to check this is probably along the lines of your
get_query_hdl_per_system function, just faster because it's all
implemented in a single system call.  And it's probably still
implemented this way.

Oh well.  Thanks for testing!


Corinna