This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
RE: overhead of reflection info
- From: "Boehm, Hans" <hans dot boehm at hp dot com>
- To: "'Adam Megacz'" <adam at megacz dot com>
- Cc: java at gcc dot gnu dot org
- Date: Fri, 20 Feb 2004 09:43:39 -0800
- Subject: RE: overhead of reflection info
You might want to look at _Jv_MarkObj, which is the backup GC mark routine.
It can still get called in relatively rare cases in which the compiler
generated mark descriptor would have been too big to fit. It needs some
of the reflection data.
Hans
> -----Original Message-----
> From: java-owner@gcc.gnu.org
> [mailto:java-owner@gcc.gnu.org]On Behalf Of
> Adam Megacz
> Sent: Thursday, February 19, 2004 7:52 PM
> To: java@gcc.gnu.org
> Subject: overhead of reflection info
>
>
>
> For future reference, I commented out make_class_data() in class.c,
> which (if I understand correctly) will inhibit the output of
> reflection data. My compressed binary size (using the UPX executable
> compressor on Linux/i86) plummeted from 728kb to 490kb. Roughly 33%
> smaller. Also, 120kb of that (post-compression) is DWARF2 info,
> meaning that 370kb should be "possible" with an unwinder specialized
> to Java that trades away runtime performance for smaller executables.
>
> I think I'm going to see how far I can get with a patch that disables
> reflection. Getting libgcj to live without it will be a challenge,
> but I think it can be done.
>
> - a
>