Request for review (#4) 7174978: NPG: Fix bactrace builder for class redefinition

[David Holmes]

Hi Coleen,

A couple of follow ups ...

On 16/01/2013 7:12 AM, Coleen Phillimore wrote:
>> Copyrights to 2013 :)
>>
>
> It's unclear whether we have to do this. I think there is going to be
> some RE process that fixes these files.

I'll believe it when I see it :)

>> src/share/vm/oops/constantPool.hpp
>>
>> ! void add_version(int version) {
>> ! if (version == -1) {
>> ! _saved._version = version; // keep overflow
>> ! } else {
>> ! int new_version = version++;
>> ! _saved._version = version == INT_MAX ? -1 : version;
>> ! }
>> ! }
>>
>> new_version is unused here but I think you meant to do
>>
>> _saved._version = new_version == INT_MAX ? -1 : version;
>>
>
> This was a pre-code-review cleanup... I didn't mean to assign version++
> into new_version. The second line stays the same. I retested this fix.

I'm unclear what the final form of this code now looks like.

> This code is only called for exceptions thrown at initialization time.
> It's hard to test this code since most exceptions during initialization
> in universe_post_init() cause
>
> # Internal Error
> (/java/east/u1/cphillim/hg/rt.bt2-clean/src/share/vm/interpreter/linkResolver.cpp:902),
> pid=10369, tid=47170653824768
> # assert(resolved_method->method_holder()->is_linked()) failed: must be
> linked
> #
>
> Should be another CR, which I thought we'd already fixed!

We did :) But we need to fix that error message so it tells us which 
class has the problem (assert really needs to take a format string as an 
arg :( ). We fixed the issue with java.lang.Class not being linked when 
trying to construct a Throwable. Maybe it is Throwable or one of the 
other exception types that is the problem.

Cheers,
David