Cleanups to Verifier.

Cleanups to Verifier.

This CL broadens the checks done by the verifier in untyped mode and introduces some subroutines to shorten the code a bit.

Introduce routines CheckUpperIs() CheckUpperMaybe() and CheckValueInputIs() that are called unconditionally by the verifier. If the typing mode is untyped, then don't check anything.

Also added a couple checks for Merge and Loop nodes that catch bugs where the operator and the node disagree on input counts (a bug encountered today).

R=mstarzinger@chromium.org, rossberg@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=24809

[Michael Starzinger, 2014-10-22 14:00:03]

mstarzinger@chromium.org changed reviewers:
+ mstarzinger@chromium.org

[Michael Starzinger, 2014-10-22 14:00:03]

LGTM (rubber-stamped).

[rossberg, 2014-10-22 14:07:36]

LGTM. My only concern is that the check indirections make it impossible to see
the origin of the failure directly, without a (working) stack trace (at least I
have repeatedly been bitten by this annoyance in other places).

[titzer, 2014-10-22 14:09:05]

On 2014/10/22 14:07:36, rossberg wrote:
> LGTM. My only concern is that the check indirections make it impossible to see
> the origin of the failure directly, without a (working) stack trace (at least
I
> have repeatedly been bitten by this annoyance in other places).

I could add more useful V8_Fatal() error messages like:

Error: node #53:Int32Add should have upper bound of Type x.

[rossberg, 2014-10-22 14:10:40]

On 2014/10/22 14:09:05, titzer wrote:
> On 2014/10/22 14:07:36, rossberg wrote:
> > LGTM. My only concern is that the check indirections make it impossible to
see
> > the origin of the failure directly, without a (working) stack trace (at
least
> I
> > have repeatedly been bitten by this annoyance in other places).
> 
> I could add more useful V8_Fatal() error messages like:
> 
> Error: node #53:Int32Add should have upper bound of Type x.

Yes, that would be great!

[titzer, 2014-10-22 14:39:50]

Committed patchset #1 manually as 24809 (presubmit successful).

[titzer, 2014-10-22 14:40:35]

On 2014/10/22 14:10:40, rossberg wrote:
> On 2014/10/22 14:09:05, titzer wrote:
> > On 2014/10/22 14:07:36, rossberg wrote:
> > > LGTM. My only concern is that the check indirections make it impossible to
> see
> > > the origin of the failure directly, without a (working) stack trace (at
> least
> > I
> > > have repeatedly been bitten by this annoyance in other places).
> > 
> > I could add more useful V8_Fatal() error messages like:
> > 
> > Error: node #53:Int32Add should have upper bound of Type x.
> 
> Yes, that would be great!

Done.