This is the mail archive of the cygwin 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]
Other format: [Raw text]

RE: Latest snapshot revamps pipe/fifo handling

Hi All...

From: Christopher Faylor Subject: Latest snapshot revamps pipe/fifo handling
Date: Sat, 7 Jul 2007 13:55:47 -0400

I've checked in a fairly major change to cygwin today.

I yanked out a lot of the win9x specific pipe handling code in the
latest snapshot and used pure Windows NT (and later) functionality
instead.  That means that fifos are actually named pipes under the hood
and both fifos and pipes are using overlapped I/O to allow themselves to
be interruptible; meaning that the special pipe thread kludge implemented
for Windows 9x is now gone.

What this means in practice is that pipe handling in snapshots may be a
little faster.  Pipes and fifos will probably also be a little less
stable since I'm sure to discover that some of the stuff I yanked out
probably had meaning even on NT-class systems.

I'll be doing more testing over the weekend but I wanted to give people
a heads up to expect some potential instability.  If you do detect a
problem please report it here, along with details.

If you don't have time to report details, need to mention how busy you
are, need to stress how much you don't know C, want to inform on how
well it used to work in previous releases, or express puzzlement that it
works fine on linux, then don't bother sending mail.  Otherwise, I'd
like to hear about problems.

Reports will include cygcheck output and the exact sequence of steps
needed to reproduce the problem.  The "exact sequence" will show the
exact commands or source code which caused the problem and will not be a
general, vague description of the problem.  The problem will not include
words like "doesn't work for me" or "crapped out" or "bit the big one"
but will instead include words like "died with signal N and here is the
stack trace" or "did not seem to provide any output as you can see from
this screen capture".

I tried out the July 8, 2007 snapshot cygwin1.dll and had the following issue (I replaced just the cygwin1.dll file).

I typed ssh localhost and got as far as the login banner and then it hung there. The bash process started by sshd was consuming all avaliable CPU. I got the same result if I used ssh to login from another machine. When I tried two simultaneous ssh connections, the created bash processes seemed to share the CPU equally (by observing them in the taks manager). I use ssh agent forwarding in case that is relevant.

The attached cygcheck output is from after I used setup to restore my cygwin1.dll. The only change was the cygwin1.dll file.


See for more details and be sure to read
more than just the part that says to include cygcheck output.

(who is now placing bets)

Unsubscribe info:
Problem reports:


Attachment: cygcheck.txt
Description: Text document

Unsubscribe info:
Problem reports:

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