Re: [rfc] rewrite of local equivalences

[canqun at nudt dot edu dot cn]

Richard Henderson <rth@twiddle.net>:

> While staring at output for the nearest intrinsic 
test case, I
> noticed how ugly the generated code for equivalences 
is.  And 
> that the existing implementation can in fact fail to 
provide
> sufficient alignment for the types involved.
> 
> My replacement implementation uses a UNION_TYPE, with 
the caveat
> that offsets of the fields are not necessarily zero.  
(Jason, do
> you think this will cause problems?  It doesn't seem 
to...)  We
> then treat the members of the equivalence more or 
less like 
> members of a C++ anonymous union.
> 
> I'm somewhat concerned that there are almost no test 
cases for
> equivalence in the test suite, so I have no idea what 
happens
> here for corner cases.
> 
> This is not a complete patch.  I have Hordes of other 
things in
> my tree at present, and I didn't want to confuse the 
issue, or
> figure out what bits this actually relies on.  It may 
be only
> exporting update_alignment_for_field from stor-
layout.c.
> 
> Thoughts?
> 

I also rewrote the implementation of COMMON blocks  by 
taking the same way. It can pass my test programs. But, 
we still need to find a way to deal with CHARACTER 
inside COMMON block. Is it worth to do?

As the code for constructing the 'UNION_TYPE' in 'trans-
equivalence.c' is very similar to which in 'trans-
common.c', also, 'trans-common.c' needs handle global 
EQUIVALENCE. Is it necessary to combine the two files?


Canqun Yang