[11u] RFR 8206895: aarch64: rework error-prone cmp instuction

[Andrew Haley]

On 3/24/21 4:33 PM, Andrew Dinn wrote:
> On 24/03/2021 14:15, Dmitry Chuyko wrote:
>> Hi,
>>
>> On 3/24/21 1:38 PM, Andrew Haley wrote:
>>> .On 3/23/21 6:29 PM, Hohensee, Paul wrote:
>>>> I'd say this patch is worth doing on its own, since it catches actual 
>>>> argument size mismatches that would otherwise be silently accepted. 
>>>> Doesn't seem super intrusive either: all it does is add formal 
>>>> argument casts to "u1", and replace some cmp instructions with subs. 
>>>> Effectively syntax-only changes, so low risk.
>>> Hmm, interesting. But it's *not* syntax only, as I mention later.
>>>
>>> OK, but it's a lot of churn in an old maintenance release,
>>> especially where there's no actual bug. Isn't churn a bad thing
>>> of itself?
>> I listed few issues that were already backported touching same code 
>> areas. And I suppose the chance of getting more of them is high right 
>> because this is a distributed change for a widely used unstruction. 
>> There is no situation in currently backported code where we would rely 
>> on non-backported argument checks though but it is an annoying thing to 
>> track.
> 
> And in almost all of these cases sorting out the patch in the absence of 
> this backport is pretty trivial. In a few cases it is not so trivial but 
> those are *exactly* the cases where backporting this patch is also 
> taking a risk.
> 
> So, this patch is not actually reducing risk. It is just importing that 
> risk into the backports repo before we need to.

We're going to have to import the MacOS and Windows AArch64 ports into 11u.
(At least I think we are. I don't think we can leave those systems without
an OpenJDK 11 port forever, can we?)

So, let's hold off to see if we're going to need 8206895 for that.

-- 
Andrew Haley  (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671