[Java PATCH] An attempt at removing a "Gross hack"

[Andrew Haley]

Steven Bosscher writes:
 > Op ma 21-07-2003, om 18:23 schreef Jeff Sturm:
 > > On Mon, 21 Jul 2003, S. Bosscher wrote:
 > > > Both patches fail miserably with Java,
 > > > which apparently sometimes uses rtl inlining, and sometimes tree inlining.
 > > 
 > > ...right, the bytecode frontend still uses rtl inlining.  Plus, a single
 > > invocation of jc1 may need both, since it's legal to compile a class file
 > > together with java source at once.
 > > 
 > > I'm also readying a patch for unit-at-a-time which might add a wrinkle.  I
 > > believe Andrew's working on converting the bytecode reader to
 > > functions-as-trees.  That might be worth having for 3.4.
 > 
 > I believe it is a must-have if we're going to fix bug 7257 properly
 > before 3.4.  Going with the langhook when the lang uses both inliners is
 > just not going to work I'm afraid...
 > 
 > For tree-ssa, the bytecode compiler already did functions as trees,
 > right?

Not already, no.  I'm working on it.

 > So it should be feasible for 3.4...  Maybe you can open a bug
 > for this to track the issue, and make 7257 depend on it?

I suppose it might be worth back porting functions as trees to
mainline, but it wouldn't be trivial.

If needs be we can diable the RTL inliner when compiling Java.  I
don't think this is a huge bug deal.

Andrew.