Re: Patch: automake 1.9 update for libgcj

[Michael Koch <konqueror at gmx dot de>]

Am Donnerstag, 5. August 2004 21:35 schrieb Bryce McKinlay:
> Michael Koch wrote:
> >-----BEGIN PGP SIGNED MESSAGE-----
> >Hash: SHA1
> >
> >Hi list,
> >
> >
> >I work a little bit with the automake update patch Tom send me and
> > its now working as expected (at least I hope so). So send this
> > patch here for testing on various platforms. On my own
> > i686-pc-linux-gnu (Debian testing) it bootstraps libgcj and works
> > successfully. It would be great to get some feedback from other
> > archs before committing.
> >
> >A patch with only the edited files is attached for easier review.
> > A full patch with regenerated files (about 1.5 MB) can be found
> > here:
>
> Cool! A few comments:
>
> 1. I noticed that the generated Makefile.in is a lot bigger than
> the current one (1.2MB vs about 250kb). I'm a bit worried about
> this particularly if it makes "make" run any slower than it already
> does. But, perhaps it is actually better now due to less dependency
> files having to be read - any ideas?

Well, this is much better then with automake 1.8. With it Makefile.in 
was around 10 MB. I think 1.2 MB is acceptable for now and we can see 
later how to improve this more after the transition is done.

> 2. Some GCJ flags (eg -Wno-deprecated) are no longer passed?

Fixed in the next iteration of the patch.

> 3. I got this error with your patch building on x86-64:
>
> /home/mckinlay/cvs/gcc-clean/build/gcc/gcj
> -B/home/mckinlay/cvs/gcc-clean/build/gcc/
> -B/home/mckinlay/gcc-x86_64/x86_64-unknown-linux-gnu/bin/
> -B/home/mckinlay/gcc-x86_64/x86_64-unknown-linux-gnu/lib/ -isystem
> /home/mckinlay/gcc-x86_64/x86_64-unknown-linux-gnu/include -isystem
> /home/mckinlay/gcc-x86_64/x86_64-unknown-linux-gnu/sys-include -g
> -O2 -C -g -classpath '' -bootclasspath
> /home/mckinlay/cvs/gcc-clean/build/x86_64-unknown-linux-gnu/libjava
>:../../../libjava -d
> /home/mckinlay/cvs/gcc-clean/build/x86_64-unknown-linux-gnu/libjava
> ../../../libjava/java/net/InetAddress.java
> ../../../libjava/java/net/InetAddress.java: In class
> `java.net.InetAddress': ../../../libjava/java/net/InetAddress.java:
> In method
> `java.net.InetAddress.isMCLinkLocal()':
> ../../../libjava/java/net/InetAddress.java:266: internal compiler
> error: tree check: expected array_type or set_type, have
> pointer_type in array_ref_low_bound, at expr.c:5485
>
> Perhaps this is a compiler bug which appears only with your patch
> due to different build order? ;-(

That's interesting. I dont get this on i386. I think Andreas Tobler 
dont got this on his solaris and powerpc machines either.

Can you look into this ?


Michael