Fix Potential NullPointException in BytesRefHash initialization

[tyronecai]

Description

because of variable shadowing, an npe may produce here. detected by Intellj ide.

final Counter bytesUsed = bytesStartArray.bytesUsed();
this.bytesUsed = bytesUsed == null ? Counter.newCounter() : bytesUsed;
bytesUsed.addAndGet(hashSize * (long) Integer.BYTES);
~~~~~~

ref: #15701

[dweiss]

I wonder why we never hit this NPE... maybe it's just not possible and we should remove the code entirely and replace it with an assertion that bytesUsed != null?

[tyronecai]

I wonder why we never hit this NPE... maybe it's just not possible and we should remove the code entirely and replace it with an assertion that bytesUsed != null?

Is it necessary to assume that bytesStartArray.bytesUsed() never null ?

just a variable shadowing issue introduced by https://github.com/apache/lucene/pull/13032/changes

[mikemccand]

Whoa, nice catch, thank you @tyronecai and IntelliJ! I wonder if there is a low-false-positive (high signal to noise ratio) static checker we could turn on in Lucene's CI builds to catch this kind of lurking bug?

I think it indeed must be that all the BytesStartArrays used here have a non-null bytesUsed? In which case +1 to make it required not null (change to assert). This is technically a public API, but @lucene.internal (well, BytesRefHash is), so we are free to suddenly change API/requirement here.