Re: [dev-serveez] gh_leave ?

[Martin Grabmueller]

> From: stefan <address@hidden>
> Date: Tue, 13 Mar 2001 09:28:22 +0100 (CET)
> 
> On Tue, 13 Mar 2001, Martin Grabmueller wrote:
> 
> > > From: "Raimund 'Raimi' Jacob" <address@hidden>
> > > Date: Mon, 12 Mar 2001 18:28:10 +0100 (CET)
> > > 
> > > when using guile from serveez i have to gh_enter() so guile takes over
> > > control. i dont know if that's cool when we fork() and exec() and
> > > stuff. can i gh_leave() somehow so that guile disappears completely ?
> > 
> > Sorry, but I don't get the problem here...  guile+serveez+anything
> > disappears completely on exec().  Actually, that's the reason of
> > existence for exec().
> 
> exce() is not actually the problem, if Guile opens its files with the
> FD_CLOEXEC flag set. But we might get a problem if we want to fork (which
> is the case for the coservers).
> The actual problem occurs as follows: On Raimi's SMP machine the wait()
> for a child in the signal handler does not return until all file
> descriptors shared with other children are closed. So if guile is running
> and we fork a new coserver it gets guile's fds as well as any other
> coserver. Thus serveez hangs if one of the coservers dies.
> I do not know if this behaviour is correct (on my laptop everything is
> fine) and if it is correctly described anyway...

Guile has a procedure port-for-each, which allows to apply a procedure
to every open port.  Oh, sorry for the Scheme speak -- a port is the
file abstraction for Scheme.  So it is possible to explicitly close
all files (except for std{in,out,err}).

Example:

(port-for-each close-port)

This could be evaluated directly after the fork().

Regards,
  'martin