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

snapshots R us


On Tue, Oct 22, 2002 at 06:00:38PM -0400, Norman Vine wrote:
>Christopher Faylor wrote:
>
>> On Mon, Oct 21, 2002 at 03:03:13PM -0400, Norman Vine wrote:
>> >Christopher Faylor writes:
>> >
>> >> I was able to duplicate Jason's mmap problem with his version of exim
>> >> and, so, I think I was also able to fix it.
>>
>> >> So, try a snapshot.  Collect them all.  Win valuable prizes.
> >
>> >rxvt seems to have a problem with this dll
>>
>> >make[1]: Leaving directory `/src/cygwin2/obj/libiberty'
>> >   2194 [main] ? 2076 open_shared: relocating shared object shared(3)
>from
>> >0xA000000 to 0xC5D0000 on Windows NT
>> >Signal 11
>>
>>The above message is a warning.  The signal 11 is a problem.  There
>>should be a stackdump file.  Please decode the addresses with addr2line
>>and report them here.
>
>It appears as if todays patch < see below > fixes this problem

That was my first stab at fixing the problem but it looks like all of
my activity in the last couple of weeks to stabilize 1.3.13 has been a
shell game.

I tried to introduce a speedup for select() in 1.3.13 by allocating a
reusable thread pool that select could use rather than stopping and
starting threads every time certain types of fds are being selected.
This worked ok except in the case of fork.  When the threads are being
initialized, Windows allocated space for their stacks.  Sometimes the
stacks showed up in places, like the heap, or in the locations where
dlls were supposed to live.  So, cygwin's fork hack got confused.  You
can't fault windows for placing the stacks whereever it feels like so,
in hindsight, the thread pool idea was doomed to failure.

I finally bit the bullet today and scrapped the "allocate a thread pool
at process start" plan and changed it to "allocate threads as they are
needed and reuse them".  If my analysis of the situation is correct, then
this should squash most of the problems with heap, mmap, and dlls that
have plagued 1.3.13 and snapshots.

So, the latest snapshot should be better.  It also has some ntsec fixes
courtesy of Pierre Humblet which should make the "incomplete /etc/passwd
file" problem a little better.  For most people, in fact, that might be
the most important change in this snapshot.

So, as usual, please try the 2002-10-22 snapshot.

cgf

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/


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