Re: Fix a tcb crash and a potential bug on mainline

[Daniel Berlin <dberlin at dberlin dot org>]

On Oct 19, 2004, at 6:30 PM, Jeffrey A Law wrote:

> On Tue, 2004-10-19 at 16:23, Daniel Berlin wrote:
>
> > Only if your pass performs *all* insertions and updates to statements
> > before performing *any* removals of statements.
> True.  However, I believe that is the case for most of our optimizers
> right now.

Except for things that just want t

> > I believe some passes don't do this right now.
> PRE and jump threading being the obvious exceptions.
PRE actually does all insertions before it does any 
eliminations/replacements these days, instead of mixing them together 
like it used to.


> > Otherwise, in the intermediate step, we would think some statements 
> > are
> > dead when they are only dead because we haven't fixed them up yet :)
> Right -- which is one of my big concerns with integrated dead
> code elimination schemes.  This kind of problem bit us with
> delete_insn as well.

I would seriously suggest we not go down the route of integrating DCE 
into immediate use replacement, of course.
Among other things, it is incredibly surprising that removing a use 
would magically make other statements disappear, unless you had 
explicitly called some DCE pass.

> jeff