[llvm] r209007 - Fix most of PR10367.

Rafael Espíndola rafael.espindola at gmail.com
Sun May 18 07:31:01 PDT 2014


> We strongly disagree on this design, and that’s fine.  But I do not see why it matters, because you not only haven’t proposed it, but you haven’t even fleshed it out to yourself yet, which is, again, fine as far as your proposal for ConstantExpr goes.  The problem is that you and Rafael have decided to go ahead and use the basic thrust of this less-than-half-baked proposal to completely redesign an existing LLVM feature in a way that intentionally constrains and complicates future extensions.  That’s using trunk as an experimental lab for IR design, and it seems totally inappropriate to me.

Sorry, what?

I removed a "feature" that would not survive llvm-as + llvm-dis. I did
so following a design that was documented in bugzilla and discussed in
the dev meeting. I explicitly talked with both Chris and Nick Kledzik
to make sure I would understand where we wanted go.

I committed the patch after having it reviewed by a Google and an
Apple employee.

This is the first time I see any objection to it and in summary it
sounds like: I may some day want to change this is some unspecified
way.

So, if you some day have some feature you want added to GlobalAlias, I
propose you do the same: Open a bug with what is needed. Discuss it
with others and send patches for review with the changes you need.

Cheers,
Rafael




More information about the llvm-commits mailing list