Uncovering Weaknesses in Neural Code Generation

TL;DR

This paper systematically evaluates state-of-the-art neural code generation models, identifying key weaknesses such as prompt inaccuracies, missing semantics, and API usage issues, to guide future research improvements.

Contribution

It provides the first comprehensive taxonomy of weaknesses in neural code generation, analyzing multiple models across diverse datasets with detailed thematic insights.

Findings

Large models fail in 26.84% of cases due to inaccurate prompts

Missing key semantics occurs in over 65% of tasks across datasets

All models struggle with proper API usage, especially with vague prompts

Abstract

Code generation, the task of producing source code from prompts, has seen significant advancements with the advent of pre-trained large language models (PLMs). Despite these achievements, there lacks a comprehensive taxonomy of weaknesses about the benchmark and the generated code, which risks the community's focus on known issues at the cost of under-explored areas. Our systematic study aims to fill this gap by evaluating five state-of-the-art PLMs: three larger models, CodeGen2.5 with 7 billion parameters, CodeGeeX2 with 6 billion parameters, GPT-4 Turbo, and two smaller ones, UnixCoder with 110 million parameters and CodeT5 base with 220 million parameters, across three popular datasets, CoNaLa, HumanEval Plus, and DS-1000. We assess the quality of generated code using match-based and execution-based metrics, then conduct thematic analysis to develop a taxonomy of nine types of weaknesses. We dissected weakness distributions in both larger and smaller models, applying an extensive methodology that encompasses model-specific as well as collective analysis (union and intersection) across models. Our research uncovers three salient findings: 1. In the CoNaLa dataset, inaccurate prompts are a notable problem, causing all large models to fail in 26.84% of cases, with even higher failure rates of 40% for smaller models; 2. Missing pivotal semantics is a pervasive issue across benchmarks, with one or more large models omitting key semantics in 65.78% of CoNaLa tasks, and similarly high occurrences in HumanEval Plus (66.09%) and DS-1000 (80.51%); 3. All models struggle with proper API usage, a challenge amplified by vague or complex prompts. Our findings aim to steer researchers towards addressing specific weaknesses and challenges in code generation. Furthermore, our annotations can offer a targeted benchmark subset for detailed analysis.