statically linked executable size
Bryce McKinlay
mckinlay@redhat.com
Thu Jul 6 04:20:00 GMT 2006
Per Bothner wrote:
> The idea that users should have to specify which libraries to link
> with is something I find totally lame.
I agree completely, and for me this is a show-stopper to splitting the
library until the linking can be made transparent to the user. The good
news is that it would be transparent once we switch to the BC-ABI by
default - any dependent classes would be found and loaded automatically
by the db mechanism. But, we're not there yet.
Splitting AWT, etc, out would likely have some effect on startup times
for simple apps, but it isn't clear how significant that would be. Most
large apps probably reference a few AWT classes anyway so they would see
little benefit. We do know that a lot of startup time is spent in the
dynamic linker. But, I think it would be more effective in the general
case to implement pointer-free metadata and build libgcj itself with
-findirect-classes, reducing reloc processing and the amount of
copy-on-write-ed .so data.
Bryce
More information about the Java
mailing list