Large language models are widely used for code generation, yet they rely on an implicit assumption that the task descriptions are sufficiently detailed and well-formed. However, in practice, users may provide defective descriptions, which can have a strong effect on code correctness. To address this issue, we conduct a large-scale empirical analysis evaluating the robustness of 10 code LLMs across 3 benchmarks under three natural language defect types: Lexical Vagueness (LV), Under-Specification (US), and Syntax/Formatting (SF). Our results show that defect impact is not uniform: Under-Specification causes Pass@1 drops of up to 15.3% with no model immune, including state-of-the-art reasoning models; Lexical Vagueness produces moderate, benchmark-dependent degradation; and surface formatting noise has a negligible effect. Crucially, robustness is shaped more by the properties of the task description than by the size or architecture of the model, in contrast to prior perturbation studies reporting uniform degradation across models. Second, to address these defects before they reach the model, we develop SpecValidator, a lightweight LoRA-fine-tuned classifier based on a 1.5B code model, achieving F1 = 0.84 and MCC = 0.79, significantly outperforming few-shot GPT-5-mini (F1 = 0.43) and Claude-Sonnet-4 (F1 = 0.56), and demonstrating that strong code generation ability does not transfer to prompt meta-reasoning. Perhaps more importantly, our analysis indicates that SpecValidator can generalize to unseen issues and detect unknown Under-Specification defects in the original (real) descriptions of the benchmarks used.

General-purpose code embeddings power tools for code search, classification, and retrieval. Compact transformer encoders for code typically rely on either human-written docstrings (labor-intensive and inconsistent) or mined structural signals such as execution traces (setting-specific and costly to collect). We empirically study an alternative: contrastive pretraining of small encoders with synthetically generated natural-language descriptions emphasizing code functionality and intent, paired with code in a dual-encoder framework at training and discarded at inference. We benchmark this approach against pretraining-based baselines, generalist LLMs, and embedding-specific models on eight retrieval, classification, and generation tasks across C, C++, and Java. Synthetic semantic supervision yields statistically significant gains over pretraining baselines of the same inference-time size on five of eight tasks, with parity on two more; once fine-tuned, it matches or exceeds zero-shot models two orders of magnitude larger on classification, and it stays on par with execution-aware supervision at matched pretraining data, suggesting a scalable, effective alternative to existing code-representation paradigms.

Grammar-constrained decoding helps large language models produce syntactically valid structured outputs, such as code, JSON, and SQL. For context-free grammars, many practical decoders enforce local prefix feasibility: each token must keep the current prefix extendable to some valid completion. Yet, under tokenizer–grammar mismatch and finite token budgets, feasible prefixes may still fail to reach acceptance. We propose a lookahead-guided decoding framework for context-free grammars based on pushdown automata. Offline, we compute bounded pushdown summaries with reachability labels and upper-bound distances to acceptance. Online, these estimates guide horizon-aware pruning and beam search. The resulting decoder is syntactically sound: every output is accepted by the target grammar. Experiments on JSON, SQL, and Linear Temporal Logic (LTL) show both consistent syntactic validity and improved completion quality over existing baselines.