Re: Eliminating Assembler Already Defined Messages for Java Library

[Per Bothner <per at bothner dot com>]

(Mark, I'm cc'ing you because your name is attached to ChangeLog
entries that seem related to this problem, and otherwise I'm afraid
you might skip the message.)

kenner@vlsi1.ultra.nyu.edu (Richard Kenner) writes:

>     If you want your patch P reinstalled, please investigate what goes
>     wrong when it is included, fix that problem, and then submit P and the
>     fix.  In the meantime, we can bootstrap.
> 
> I'll be glad to do that and I indeed already offered.  Please show me the
> code that gets miscompiled and we'll see what's going wrong.

The code that gets mis-compiled is the Java run-time library
when you do a bootsrap under ix86 Linux.

I've looked a little more at the problem.  I am willing to believe
it may be a problem in the Java front-end, but I don't really have
a good handle on the problem or how to fix it.

The problem involves a block created by expand_fixup.  This needs
to be inserted somehow into the block structure - but where?
Some complicating factors:

* cfun->x_whole_function_mode_p is not set.  It perhaps should be
(when compiling from source).  However, the Java front-end pre-dates
x_whole_function_mode_p.  Just setting x_whole_function_mode_p = 1
in the front-end seems plausible, but triggers other problems.

* jc1 uses a pun where the BLOCK_SUBBLOCKS field of a block is
used for the BLOCK_EXPR_BODY of a block (i.e. the "body"
expression of a block, typically a OCMPOUND_EXPR).  The BLOCK_EXPR_BODY
is used after parsing and up until expand_expr; BLOCK_SUBBLOCKS field
needs to be correct at final.  I don't know if this is an issue,
but it could be a contributing complicating factor.

It might be cleaner to replace BLOCK with a BIND_EXPR, but it might
be better if we could defer that until after 3.0.

So before we try to hack together a solution, we need to understand
/ agree on some things:

(1) is setting x_whole_function_mode_p to 1 the Right Think to do?

(2) Given that we have a tree of BLOCK nodes created by the front
end, when expand_fixup creates a block, where should it go?  The
code for when cfun->x_whole_function_mode_p is true is this:
	    BLOCK_CHAIN (block)
	      = BLOCK_CHAIN (DECL_INITIAL (current_function_decl));
	    BLOCK_CHAIN (DECL_INITIAL (current_function_decl))
	      = block;
This doesn't work, because DECL_INITIAL hasn't been set yet,
but that can be fixed.  However, my attempts at do so haven't
worked, so far.  If we leave x_whole_function_mode_p false,
then insert_block gets called - and seems to do the wrong thing.
Again, that can probably be fixed - if I knew where the new
block should go.
-- 
	--Per Bothner
per@bothner.com   http://www.bothner.com/~per/