Pronoun Coreference Resolution (PCR) is the task of resolving pronominal expressions to all mentions they refer to. 
The correct resolution of pronouns typically involves the complex inference over both linguistic knowledge and commonsense knowledge.
Recently, with the help of pre-trained language representation models, the community has made significant progress on several dataset including CoNLL-2012 and winograd Schema Challenge (WSC).
However, as most existing works focus on developing PCR models for specific datasets, it is still unclear whether current PCR systems are reliable in real applications.
Motivated by this, we propose a new PCR benchmark PCR4All, which evaluates PCR systems from different angles (i.e., knowledge resource, domain, frequency, distance).
Experiments demonstrate that there is still a notable gap between existing PCR models and a reliable PCR system because they are often trained to optimize a specific kind of PCR while ignoring others.
We hope that PCR4ALL can motivate the community to pay more attention solving the overall PCR problem rather than training a model towards a specific dataset.