In today's widely diffused CSMA wireless networks, problems of coordination in the access to the channel by multiple transmitters can lead to unfair situations where some of the flows receive much of the network throughput while others suffer from poor performance. As a solution to this problem, recent theoretical studies have proposed distributed CSMA adaptation protocols that, under certain assumptions, maximize a network utility function, yielding high throughput fairly distributed among flows. The main idea in the operation of such protocols, referred to as Optimal CSMA, is to adapt the contention aggressiveness of a flow as a function of its queue length, without the need of any information exchange among nodes. Thus, their operation is distributed, and does not introduce additional control overhead to CSMA operation. However, we show that such an approach is fragile, and can suffer high performance degradation under conditions of frequent occurrence, namely; asymmetric channels, heterogeneous traffic, and packet collisions. In this work, we address the main sources of performance degradation in Optimal CSMA to design a distributed protocol for proportional-fair throughput maximization robust to such conditions. First, we generalize Optimal CSMA models to incorporate individual per-link modulation and coding rates. With our generalized network optimization model, we derive distributed algorithms that maximize utility under arbitrary channel capacities. Second, we propose a novel structure that can be used in the place of queues to provide optimal CSMA adaptation. As such a structure does not use traffic backlog to operate, the resulting adaptation is optimal for the set of active flows under general traffic arrival patterns. Third, we propose a robustness function compatible with the optimization approach, which maintains high medium access rates to maximize performance in low contention scenarios, yet reduces medium access to avoid collisions as the network contention increases. Finally, we validate our design by evaluating its performance against state-of-the-art protocols for distributed CSMA optimization under critical scenarios combining the three aforementioned sources of performance degradation, observing vast gains in network logarithmic utility across a wide-range of network operating conditions.