Curve Ball for Rare Classes

You're comparing two classifiers on a dataset where only 2% of examples belong to the positive class. Why would a practitioner generally prefer a precision-recall (PR) curve over a ROC curve to evaluate performance on the rare positive class?

ROC curves require the positive and negative classes to be equal in size by definition

Precision depends only on true positives and false positives, so a PR curve exposes wrong positive predictions directly, whereas the ROC curve's false positive rate is diluted by the huge pool of true negatives and can look optimistic under heavy imbalance

ROC curves cannot be computed when classes are imbalanced

PR curves and ROC curves always produce identical rankings of classifiers, so the choice never matters