Multi-view subspace clustering (MSC) has attracted significant attention due to its superiority in unsupervised multi-source data analysis. For MSC methods, the key is to explore the local view-specific and global view-shared information embedded in multi-view data. However, most existing MSC methods only explore view-specific information or shared information across multi-view data, ignoring the interaction between global and local information. This paper proposes a Comprehensive Multi-view Subspace Clustering (CMSC) with global-and-local representation collaborative learning. Specially, view-specific subspace representations are learned in the original multi-view feature space for exploring complementary local description information. And, view-shared subspace representation learned in shared latent space could explore global consistency among multi-view data. To achieve the interaction of global-and-local information and fully explore the complex correlation among multi-views, local view-specific and global view-shared representations are uniformly merged into a low-rank tensor regularized by non-convex and more compact tensor logarithmic nuclear norm. Thus, the proposed CMSC can sufficiently explore the complementarity-consistency among multi-view data and capture more comprehensive semantic correlation. Experimental results on several benchmark datasets verify the performance superiority of CMSC over some classic MSC methods.