Re: Do you need the vtable bug fixed?

[mark at codesourcery dot com]

>>>>> "Martin" == Martin v Loewis <martin@mira.isdn.cs.tu-berlin.de> writes:

    Martin> Well, there are two steps in this involved. First, the
    Martin> code must get into the compiler. Then, it can be enabled
    Martin> by default.

I don't think this belongs in gcc 2.95.  We're at the point where
we're making very, very small changes in order to get to a stable
release.  I agree; your changes will not change behavior in the
default case.  But, I have been wrong before.  It is very, very hard
to know just what might break, even when it looks like nothing could
change.

For example, even code that is never executed could cause GCC's
optimizer to miscompile neighboring code, resulting in a cc1plus that
generates bad output, where without your code it would generate good
output.  That's not the fault of your code, but it's not a risk we
should be taking.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com