remove_unnecessary_notes breaks libjava build

Per Bothner per@bothner.com
Tue Mar 6 08:03:00 GMT 2001


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

> But the only time BLOCK_DEAD is set is when the notes are deleted, so I'm
> confused.

Well, it seems to be related to exception handling.  My guess (without
knowing what is going on) is that exception handling is doing things
with instruction sequences that are confusing things.

> Can you look a bit more into this?

I really think you should fix this.  Not only did you make the change
which broke things, but also this is involving insns and exception
handling, things you understand better than I.

It should be easy enough to reproduce the problem.  I did:

 /home/bothner/GNU/gcc/configure --enable-threads --prefix=/home/bothner/GNU/linux --enable-languages=c++,java

I then did a 'make boostrap'.  After a while, the make fails.
This was on a i686-pc-linux-gnu (running Red Hat 7).

The same problem has been reported by other people.

>     The quick work-around is basically disabling the patch:
> 
> Which will cause the abort in function.c to be hit.

Since I don't know which of the many aborts in function.c you are
talking about, this illustrates that I am not the person to track
this down.
-- 
	--Per Bothner
per@bothner.com   http://www.bothner.com/~per/



More information about the Gcc-bugs mailing list