Résumé : TLS is an end-to-end protocol designed to provide confidentiality and integrity guarantees that improve end-user security and privacy. While TLS helps defend against pervasive surveillance of intercepted unencrypted traffic, it also hinders several common beneficial operations typically performed by middleboxes on the network traffic. Consequently, various methods have been proposed that "bypass" the confidentiality goals of TLS by playing with keys and certificates essentially in a man-in-the-middle solution, as well as new proposals that extend the protocol to accommodate third parties, delegation schemes to trusted middleboxes, and fine-grained control and verification mechanisms. We first review the use cases expecting plain HTTP traffic and discuss the extent to which TLS hinders these operations. We retain 19 scenarios where access to unencrypted traffic is still relevant and evaluate the incentives of the stakeholders involved. Second, we survey 30 schemes by which TLS no longer delivers end-to-end security, and by which the notion of an "end" changes, including caching middleboxes such as Content Delivery Networks. Finally, we compare each scheme based on deployability and security characteristics, and evaluate their compatibility with the stakeholders' incentives. Our analysis leads to a number of key findings, observations, and research questions that we believe will be of interest to practitioners, policy makers and researchers.