[lworld] RFR: 8367242: [lworld] C2 compilation asserts with "dead loop detected" [v3]

[Marc Chevalier]

On Wed, 1 Oct 2025 05:40:25 GMT, Tobias Hartmann <thartmann at openjdk.org> wrote:

>> I wouldn't fight too hard for it, but when trying things, it was necessary (since I wanted different types for `uin`), but I also think it helps to see that it is not useful too long, and I can mess with it without having to look at what comes after. In this case, indeed, it feels a bit coming out of nowhere at the end...
>> 
>> In C++17, I'd do:
>> ```c++
>>  if (Node* uin = unique_constant_input_recursive(phase); uin != nullptr) {
>>   return uin;
>> }
>> 
>> one could also do
>> ```c++
>>  if (Node* uin = unique_constant_input_recursive(phase)) {
>>   return uin;
>> }
>> 
>> if we are not afraid of implicit conditions (but we are, and I'm fine with it!).
>
> Right but now that both `uin` have the same type, the scoping seems unnecessary.

Unnecessary to make it compile. But I still think it's helpful: each block can be rearranged since it doesn't change the semantics of an outer-living variable. It makes explicit that `uin` isn't used after... But I said I won't fight too hard, so I reverted!

-------------

PR Review Comment: https://git.openjdk.org/valhalla/pull/1640#discussion_r2393981991