Re: -fnew-verifier

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:
 > Bryce> Agreed. Also, if we arn't going to make --indirect-dispatch the
 > Bryce> default for 4.0, then we should at least make it the default on the
 > Bryce> HEAD as soon as 4.0 branches.
 > 
 > Bryce> Are there any strong against making it the default, anyway?
 > Bryce> Performance? Bugs?
 > 
 > AFAIK right now -findirect-dispatch only works when compiling from
 > .class.

I've been studying this problem.  The source compiler makes a lot of
assumptions about the exact way that trees are laid out.  When
compiling with -findirect-dispatch there is another layer of
indirection, and therefore the source compiler can't patch the trees
corectly.  It is possible to fix this, but probably not in the 4.0 time
frame.

Therefore, I suggest we simply disable -findirect-dispatch when
compiling from source.  Should we produce a warning?  If so, we need
to fix the testsuite so that we test the bytecode compiler with
-findirect-dispatch, but not the source compiler.

Andrew.