Graph contrastive learning is an important method for deep graph clustering. The existing methods first generate the graph views with stochastic augmentations and then train the network with a cross-view consistency principle. Although good performance has been achieved, we observe that the existing augmentation methods are usually random and rely on pre-defined augmentations, which is insufficient and lacks negotiation between the final clustering task. To solve the problem, we propose a novel Graph Contrastive Clustering method with the Learnable graph Data Augmentation (GCC-LDA), which is optimized completely by the neural networks. An adversarial learning mechanism is designed to keep cross-view consistency in the latent space while ensuring the diversity of augmented views. In our framework, a structure augmentor and an attribute augmentor are constructed for augmentation learning in both structure level and attribute level. To improve the reliability of the learned affinity matrix, clustering is introduced to the learning procedure and the learned affinity matrix is refined with both the high-confidence pseudo-label matrix and the cross-view sample similarity matrix. During the training procedure, to provide persistent optimization for the learned view, we design a two-stage training strategy to obtain more reliable clustering information. Extensive experimental results demonstrate the effectiveness of GCC-LDA on six benchmark datasets.