[Tcl-bugs] [ tcl-Bugs-2860923 ] zlib-9.3 hangs

[SourceForge.net <no...@so...>]

Bugs item #2860923, was opened at 2009-09-17 16:15
Message generated for change (Comment added) made by dgp
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=110894&aid=2860923&group_id=10894

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: 24. Channel Commands
Group: development: 8.6b1.1
Status: Open
Resolution: Works For Me
Priority: 5
Private: No
Submitted By: Don Porter (dgp)
Assigned to: Andreas Kupries (andreas_kupries)
Summary: zlib-9.3 hangs

Initial Comment:

Testing on an old IRIX box...

$ uname -a
IRIX64 xanadu 6.5 10070055 IP30

I see test zlib-9.3 hang.


----------------------------------------------------------------------

>Comment By: Don Porter (dgp)
Date: 2010-08-26 14:12

Message:

and that means this is a general problem, not just
on Solaris, just that different systems have different
limits.  A bit of experimenting finds that the max
working value on my Linux system is
172032 = 42 * 4096.

That is, a test with 172032 bytes written
to the socket passes, while a test with
172033 bytes hangs.

I'm not sure whether this test is the right
place to deal with this particular issue.
The test may just be testing other things
and the right answer is to just pick a value
small enough to not hang on all tested platforms.

We may need another test or tests to deal with
this, and maybe even more depending on how
much we want blocking [close] problems to be
solved by Tcl vs. making them the application's
problem to deal with.



----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2010-08-26 13:59

Message:

can't find docs for send() on Solaris 10, but
the docs on Linux say:

"When the message does not fit into  the  send  buffer  of  the  socket,
       send() normally blocks, unless the socket has been placed in
non-block-
       ing I/O mode."

Apparently we're stuck blocking until the reader (which is us!)
can suck up some of the blockage?

----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2010-08-26 11:45

Message:
Hangs in the 13th call to send() in
TcpOutputProc()

----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2010-08-26 11:30

Message:

49152 = 12 * 4096

----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2010-08-26 10:45

Message:
when it hangs, it hangs in [update]
before [fcopy] is ever called.


----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2010-08-26 10:16

Message:
dgp	size 49153 hangs
dgp	size 49152 does not

----------------------------------------------------------------------

Comment By: Pat Thoyts (patthoyts)
Date: 2010-08-26 10:00

Message:
zlib-9.3 does not actually test the zlib transforms. It is supposed to be a
check that the test structure used for  zlib-9.x tests is valid. It sets up
a socket with no transform added (ie: an identity transfer) and pushes some
data across. In the other tests we do the same thing but add various zlib
transforms on top.
There is a timeout in here too - so presumably this is jamming in the
asynchronous fcopy and jamming up the event loop.

----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2010-08-26 09:23

Message:

See the same problem now on a Solaris 10 box.

uname -a
SunOS entity.cam.nist.gov 5.10 Generic_142900-13 sun4u sparc
SUNW,Sun-Blade-1000


----------------------------------------------------------------------

Comment By: Donal K. Fellows (dkf)
Date: 2009-10-11 07:35

Message:
Also works fine with an unthreaded build (this is all on OSX).

----------------------------------------------------------------------

Comment By: Donal K. Fellows (dkf)
Date: 2009-10-11 07:34

Message:
Cannot duplicate (threaded build). Pat's had more luck tracking down these
sorts of problems, so assigning to him.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=110894&aid=2860923&group_id=10894