Graph Convolutional Network (GCN)-based collaborative filtering has achieved remarkable progress in recommendation systems, yet existing methods still struggle with two key issues: limited use of high-order collaborative signals and ineffective negative sampling. To address these challenges, we propose a novel framework that integrates high-order collaborative positive sampling (HCPS), hard negative sampling (HNS), and structural contrastive loss (SCL). HCPS enriches supervision by incorporating high-order neighbors as potential positives, alleviating sparsity beyond direct interactions. HNS constructs semantically close but unobserved hard negatives, guiding the model to learn fine-grained preference boundaries. SCL further enhances robustness by explicitly contrasting potential positives with hard negatives, reducing noise. Extensive experiments on multiple real-world datasets demonstrate that our framework consistently outperforms state-of-the-art baselines, confirming the effectiveness of jointly mining informative positives, constructing challenging negatives, and enforcing structural contrastive learning in recommendation tasks. Our code is available on GitHub ( https://github.com/Hannah0430/DSGCL ).