This is the mail archive of the cygwin-developers@cygwin.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: 20011101 snapshot broken (also 20011103)


On Sun, Nov 04, 2001 at 08:37:18PM -0500, Christopher Faylor wrote:
>On Mon, Nov 05, 2001 at 10:30:42AM +0900, Kazuhiro Fujieda wrote:
>>>>> On Mon,  5 Nov 2001 01:13:34 +0000
>>>>> David Starks-Browning <starksb@ebi.ac.uk> said:
>>
>>> However, with this snapshot, we are back to ssh-agent + ssh-add not working:
>>> 
>>>   $ eval `ssh-agent`
>>>   $ ssh-add
>>>   Could not open a connection to your authentication agent.
>>
>>I guess this is triggered by incorrect handling of unreliable
>>pipes on Win9x/Me. Please see the following example:
>>
>>$ echo `echo tako`
>>
>>$
>
>Oddly enough, both of the above work fine for me.
>
>If you have a clue about what is now incorrect in the handling of
>unreliable pipes, please share it.

I've made a minor change to the pipe reading code.  If someone could try
this and confirm/deny that it works, I'd appreciate it.

The specific change is in fhandler_pipe::read.  I'm changing the initial

  if (hiteof ())

to

  if (broken_pipe)

This essentially makes the test Windows NT specific and should revert
cygwin to the previous 1.3.4 behavior.

My theory is that my Windows 98 machines are not seeing this problem
because they are too fast.  I only have a 1GHZ and a 733MHZ system to
check this on.  I have a slower laptop that I could boot into Windows 95
but it is running another test right now.  ssh-agent doesn't work there
anyway so it wouldn't be a valid test.

If someone could change back and forth between those two cases and
verify if it fixes/causes the problem, it would be helpful.  I'm uploading
a new snapshot now.

cgf


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]