Re: [RFC v0 0/4] Give a type to constants too

From: Linus Torvalds <torvalds@linux-foundation.org>
To: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Cc: Sparse Mailing-list <linux-sparse@vger.kernel.org>,
	Dibyendu Majumdar <mobile@majumdar.org.uk>,
	Christopher Li <sparse@chrisli.org>,
	Jeff Garzik <jeff@garzik.org>, Pekka Enberg <penberg@kernel.org>
Subject: Re: [RFC v0 0/4] Give a type to constants too
Date: Thu, 16 Mar 2017 10:25:08 -0700	[thread overview]
Message-ID: <CA+55aFyqMScsVe8-5T=omdpSsFaYmdD-PbK3KqKW=1gXgitMaA@mail.gmail.com> (raw)
In-Reply-To: <20170311154725.87906-1-luc.vanoostenryck@gmail.com>

Sorry for not reacting to this earlier..

On Sat, Mar 11, 2017 at 7:47 AM, Luc Van Oostenryck
<luc.vanoostenryck@gmail.com> wrote:
> This is a RFC for giving a type to constants/PSEUDO_VALs.

This seems completely broken. Not from an implementation standpoint,
but from a conceptual one.

To explain, let me give a completely idiotic example:

    unsigned int test(int arg)
    {
        return arg + (unsigned int)arg;
    }

note how we're adding a "int" and an "unsigned int" together. But the
CSE etc doesn't actually care at all, and we will linearize this to
just

    test:
        add.32      %r5 <- %arg1, %arg1
        ret.32      %r5

because the type just isn't relevant at the linearization phase.

You can tell that there *used* to be multiple pseudos from the
numbering ("%r5"? What happened to "%r1..4"?), but they have all been
smushed together.

Linearization has fundamentally gotten rid of all the C types, and all
you can find are some rough remnants of them (you can find the *size*
of the type, and you can find the rough "type" of type - is it a
pointer, FP value or integer. There aren't even any signs, although
some _operations_ are signed (but not the pseudos).

The same pseudo can have many different types.

                    Linus