Re: gdb with gcj

[Andrew Haley <aph at redhat dot com>]

Bryce McKinlay writes:
 > On May 1, 2004, at 9:41 AM, Andrew Haley wrote:
 > 
 > > The last time I investigated this, some of the optimizers (CSE?
 > > loop?) were replacing one variable with another.  This doesn't
 > > change the semantics of the program but the new variable is no
 > > longer marked as a user variable and doesn't carry debug info.
 > 
 > Lets make sure there is a PR on this. We shouldn't have to recompile 
 > just to debug.

Well, it's been like that for a long while, and AFAIK it's a hard
problem.  In theory you can generate DWARF expressions that allow you
to calculate what a variable would be if the code were not optimized,
but in the general case it's impossible.  After all, the "as if" rule
allows the compiler completely to rewrite a program.

 > There is something similar already for tree-ssa:
 > 
 > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14638
 > 
 > Hopefully a solution for the tree-ssa issue will apply here as well.

The variable renaming in that testcase is very much a tree-SSA
specific thing.

In any case, until we have a real testcase it's impossible to know
what the real problem is.  It may just be a bug.

Andrew.