1.7.5: Occasional failure of CreatePipe or signal handing due to thread-unsafe code in cwdstuff::set
Wed Sep 15 13:00:00 GMT 2010
On Sep 14 22:09, John Carey wrote:
> On Sep 14 09:11, Corinna Vinschen wrote:
> > I applied the below patch to Cygwin CVS and it appears to work nicely.
> > The only potential race I can think of is, if another thread of the same
> > Cygwin process calls SetCurrentDirectory. I'm inclined to let this
> > situation slip through the cracks since SetCurrentDirectory will already
> > mess up a mixed-mode Cygwin process.
> > The "wincap.has_transactions ()" is just for testing. If we can use
> > that code, I would replace is with something like
> > "wincap.has_messed_up_cwd_handling()" or so.
> I would argue against introducing a race with threads that call
> SetCurrentDirectory(), for at least the following reasons:
> 1. The race can be avoided without too much grief.
Right. Still, the licensing issue...
> 2. If a program does violate the rule against calling
> SetCurrentDirectory() directly (possibly because of
> some third party DLL being involved), then it could be
> very difficult to detect the source of the resulting problems.
But we're talking about mixed-mode Cygwin apps. They should use the
Win32 API with care anyway. And the disconnection between
SetCurrentDirectory and chdir already exists if the latter is called at
> 3. The unknown.
Hard to argue aginst *that*...
> Anyway, one other detail:
> Are races within a pure-Cygwin program are prevented by
> "cwd_lock"? I don't see it being locked yet, just initialized.
> Perhaps the cwdstuff::set patch is not yet written?
There's a cwd_lock.acquire and a cwd_lock.release call in
cwdstuff::set. There's no such lock when cwdstuff::override_win32_cwd
is called from cwdstuff::init, because at the time init is called there's
no other concurrent thread running, except for the signal thread.
Corinna Vinschen Please, send mails regarding Cygwin to
Cygwin Project Co-Leader cygwin AT cygwin DOT com
Problem reports: http://cygwin.com/problems.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
More information about the Cygwin