Re: [cp-patches] Re: RFC: Reduce number of split-for-gcj lists

[Mark Wielaard]

On Sun, 2005-09-18 at 14:33 +0200, Mark Wielaard wrote:
> On Mon, 2005-09-12 at 03:05 +0200, Mark Wielaard wrote:
> > On Sun, 2005-09-11 at 18:48 -0600, Tom Tromey wrote:
> > > >>>>> "Mark" == Mark Wielaard <address@hidden> writes:
> > > 
> > > Mark> It does mean a change in any of the deeply nested packages triggers
> > > Mark> re-compilation of a few more files in its 'parent-package'. But I 
> > > feel
> > > Mark> the overall compile time reduction is worth it.
> > > 
> > > Does it noticeably impact memory use?  One reason we did the splitting
> > > this way rather than all-at-once was that some compilations used an
> > > excessive amount of memory.
> > 
> > Not noticeably on my machine (it does when I reduce it to just 2 levels,
> > then swing takes up a lot of memory - it does slashes compile time in
> > half again...). But my dev-machine has lots of memory. I can explicitly
> > try on one with much less memory later this week.
> 
> I tried testing on an old ppc machine with just 192MB memory, doubled to
> a whopping 384MB through the use of swap. Most of the build goes fine.
> But unfortunately our friend bug #21418 seems to be back:
> 
> /usr/bin/gcj -Wno-deprecated --encoding=UTF-8 --bootclasspath '' --classpath 
> ..:../../classpath/vm/reference:../../classpath:../../classpath/external/w3c_dom:../../classpath/external/sax:.:
>  -C -d . -MD -MF lists/javax-swing.deps -MT lists/javax-swing.stamp -MP 
> @lists/javax-swing.list
> ../../classpath/javax/swing/Box.java:67: error: Nested class 
> java.awt.Container$AccessibleAWTContainer is protected; cannot be accessed 
> from here.
>      protected class AccessibleBox extends Container.AccessibleAWTContainer
>                                            ^
> ../../classpath/javax/swing/Box.java:92: error: Nested class 
> java.awt.Component$AccessibleAWTComponent is protected; cannot be accessed 
> from here.
>          extends Component.AccessibleAWTComponent
>                  ^
> 2 errors
> 
> So my testing on this box is halted again till I can figure out another
> workaround. Sigh. None of my other setups show this bug anymore which is
> kind of strange.
> 
> I still think the patch is valuable since it saves so much compile time
> and the above error occurs with and without my patch.

With Ranjit's fix for gcj bug 21418 in I tested again on this machine.
(cd lib; time make)
Before the patch:

real    62m13.797s
user    49m15.560s
sys     10m16.950s

After the patch:

real    41m24.718s
user    29m54.690s
sys     8m44.460s

So I am going to check this in.

Cheers,

Mark

signature.asc
Description: This is a digitally signed message part