RFR: 8253286: Use expand_exact() instead of expand_at() for fixed requests in G1

[Stefan Johansson]

On Mon, 21 Sep 2020 10:27:05 GMT, Albert Mingkun Yang <ayang at openjdk.org> wrote:

>>> Sth like expand_exact(curr, (uint)-1, NULL). (Ofc, this is a contrived example.)
>> 
>> True, with bad input `expand_exact()` will likely assert or crash. But as you've noticed, we make sure that the input
>> is within bounds and valid. I agree that using a for-loop here is better, and I can file an enhancement for this. I
>> would like this change to only address how we expand the found regions.
>>> Maybe this is obvious to veterans, but making it explicit is more friendly for newcomers, IMO.
>> 
>> I agree that making the code easier to understand is great, and to me using `expand_exact()` which can only expand this
>> exact region should make it easier to follow. Even though the name doesn't say so `expand_at()` will expand any region
>> starting at the given index, which to me is both wrong and very confusing for cases where you've found a specific
>> region you want to make available.
>
>> I would like this change to only address how we expand the found regions.
> 
> Sure.
> 
> Thank you for the explanation. I think we have reached an agreement here; please update the pre-call comments and
> post-call assert as discussed above. Thank you again.

Thanks! I also filed [JDK-8253420](https://bugs.openjdk.java.net/browse/JDK-8253420) for the `find_highest_free()`
refactoring.

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

PR: https://git.openjdk.java.net/jdk/pull/235