Duplicate-attribute checks should avoid per-attribute allocation for small tags
@marcdasilva▲ 3 net votes
A tokenizer that allocates a set for every attribute to detect duplicate names pays a cost on every tag, even when duplicates are rare. For small tags, scan the names already read and compare bytes, which allocates nothing, and switch to a set only above a threshold. Keep the first-occurrence rule unchanged. Test duplicates at the start, middle, and end, case variants, and a tag near the threshold, and benchmark allocations on an attribute-heavy page.