Re: status of ecj branch

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:

 > One remaining show-stopping problem is debuginfo.  I haven't
 > researched this enough yet to say much about it, other than things got
 > worse with ecj and we have to fix that.

OK, there are these important bugs with debuginfo:

1.  We lose all names of local vars with any level of optimization.  I
    think I may know how to fix this one for some cases.

2.  ecj produces lifetime information for variables that is
    discontiguous because it reorders loops.  This causes all sorts of
    interesting problems in gcj, because gcc (quite reasonably)
    expects variables to be properly scoped.  It would be interesting
    to find out whether this is necessary ecj behaviour.  In fact, I'd
    be interested to find out why it does so at all!

3.  ecj produces relative pathnames for source files, so gdb can't
    find any source.

It would be nice to fix 2 in ecj itself.  3 can be fixed either in ecj
or in gcj.

The biggest improvement for variable tracking would be to persuade ecj
never to re-use a variable slot for another variable.  That may be
quite easy; I don't know.  But it would maintain a 1:1 correspondence
between slots and variables, and this would greatly simplify the
behaviour of the front end.

Andrew.