splitting dependencies in cygwin1.dll

Robert Collins robert.collins@itdomain.com.au
Sat May 25 20:30:00 GMT 2002



> -----Original Message-----
> From: Christopher Faylor [mailto:cgf@redhat.com] 
> Sent: Saturday, May 25, 2002 10:45 AM
> To: cygwin-developers@cygwin.com
> Subject: Re: splitting dependencies in cygwin1.dll
> 
> 
> On Fri, May 24, 2002 at 07:37:31PM +1000, Robert Collins wrote:
> >I'd like to split out some of the classes & headers in 
> cygwin1.dll to 
> >make it more loosely coupled.  This wouldn't make it less 
> able to use 
> >inlined calls, but would reduce recompiles where non-related things 
> >have not actually changed.
> >
> >I'd also like to make the headers consistently forward 
> declare required 
> >classes / include their own pre-requisite headers (with forward 
> >declarations preferred).
> >
> >I'll run all such changes through cygwin-patches for review, 
> but first: 
> >do you have any objections to this being done?
> 
> Perhaps you may recall a similar request a few months ago, 
> where someone was going to "fix" the cygwin headers.
> 
> At that time, I said that I didn't want to get into a 
> discussion of the best way to deal with headers and I 
> mentioned that I wasn't going to have headers include other 
> headers at that time.  That hasn't changed.

Ah well, I can always hope.
 
> Robert, if you're looking for things to do, why not finish 
> the sysv ipc cygserver work?  I think that would be much more 
> bang for the buck than messing with headers.

Not really. Recompiles take quite some time for me, and I hate the way
things rebuild more than they need to. The profiling effort - which is
where the header frustration really came up - is in aid of the cygserver
effort. I need to understand where the horrible latency is coming from
(and Ralf has found other areas with high latency) before solving it.
Completing the code without understanding if there is a design flaw
would be a waste of time.

So my current cygwin todo list looks like:
* get profiling working. This is very very close, but not quite right
yet.
* Profile an rpc test run with Ralf's tools.
* Analyze the results.

Rob



More information about the Cygwin-developers mailing list