[tree-ssa] Inlining vs gimple vs compound expressions

[Andrew MacLeod]

On Thu, 2003-02-27 at 16:01, law@redhat.com wrote:
> In message <1046378181.760.91.camel@steven>, Steven Bosscher writes:
>  >Op do 27-02-2003, om 20:46 schreef law@redhat.com:
>  >> 
>  >> 
>  >> Diego/Andrew -- Back when we were in Toronto, we discussed getting rid of
>  >> COMPOUND_EXPRs.  Where do we stand on this?
>  >
>  >For those who are trying to write a front end that is based on the
>  >tree-ssa work, would you please be more specific about what you mean
>  >with "getting rid of COMPOUND_EXPRs"?  Does it mean that GENERIC/GIMPLE
>  >will not use them anymore?
> Basically CEs are a rather heavy-weight means to link trees together, both
> in terms of their memory requirements and the silly requirement that
> have to strip them away anytime we want to do anything with the underlying
> node.
> 
> If you're not careful you start to end up with STRIP_foo things everywhere
> because you don't ever know if you're looking at the real node or the CE
> container.  And you still miss some, leading to weird failures because
> you were looking at the container rather than the real expression.
> 
> The basic idea was to avoid all that crazyness with a scheme that worked
> better.  However, based on messages from Andrew it doesn't sound like 
> CEs will be going away anytime soon.
> 

They might not be going away, but they will be transparent. You should
never have to look for or create a CE node explicitly. You *should* be
able to do everything through iterators and iterator manipulators. 

Thats the plan anyway :-)

Andrew