SHF: Small: Solving the Search for Relevant Code in Large Repositories with Lightweight Specifications

搜索相关代码是程序员的一项常见任务，随着开放代码库的规模不断扩大，程序员有机会重用其中的资源，这项任务对他们的工作效率变得越来越重要。然而，供程序员利用如此庞大的存储库的代码搜索工具在过去十年中几乎没有发展。程序员仍然通过指定关键字进行搜索，这种方法的执行可能非常不一致，因为它取决于程序员提供与存储库中程序的实现或文档编制方式相匹配的术语的能力。这是有问题的，因为一个稍微错误的关键字可能会错过相关的结果，并返回大部分虚假结果，降低程序员的工作效率。本研究旨在通过探索一种在两个方面新颖的代码搜索来解决这些限制。首先，它允许程序员通过以输入和输出的形式指定所需的代码行为来搜索代码。这消除了程序员想象问题的解决方案可能是如何实现的需要，取而代之的是，程序员可以专注于定义软件应该做什么。其次，给定描述期望行为的输入和输出以及代码存储库中的程序集，两者都被自动编码为约束，所提出的方法使用约束求解器来识别存储库中的哪些程序可以满足程序员的规范，从而有效地解决了搜索问题。这保证了搜索结果按照指定的方式运行。开发这种方法需要解决几个基本挑战，包括定义映射以自动高效地将程序编码为约束，以便求解器能够找到合适的解决方案。还必须开发系统的策略，以便在程序员的规范太弱并返回太多匹配时改进它们，并在没有确切匹配的情况下放松约束以找到部分匹配。支持部分匹配组合的技术也将是必要的，以将该方法扩展到更大和更多样化的搜索查询。最后，必须建造基础设施并进行研究，以确定拟议的方法在实践中是否具有成本效益。这些技术的使用将改变程序员的操作方式，使他们能够真正利用日益丰富的代码存储库中的现有代码，从而直接影响他们的生产力。

Searching for relevant code is a common task among programmers, one that is becoming more crucial to their productivity as open code repositories grow in size and programmers have the opportunity to reuse the resources therein. Yet, the code search tools for programmers to leverage such massive repositories have barely evolved in the last decade. Programmers still search by specifying keywords, an approach that may perform very inconsistently as it depends on the programmer's ability to provide terms that match how the programs in the repository are implemented or documented. This is problematic because a slightly wrong keyword could miss relevant results and return mostly spurious results, reducing programmer's productivity.This research aims to address these limitations by exploring a code search that is novel in two aspects. First, it allows programmers to search for code by specifying desired code behavior in the form of inputs and outputs. This removes the need for the programmer to imagine how the solution to a problem may have been implemented, and instead the programmer can concentrate on defining what the software should do. Second, given the inputs and outputs describing the desired behavior and the set of programs in a code repository, both automatically encoded as constraints, the proposed approach employs a constraints solver to identify what programs in the repository could satisfy the programmers' specifications, effectively solving the search. This guarantees the search results behave as specified. Developing this approach requires tackling several fundamental challenges including the definition of mappings to automatically and efficiently encode programs as constraints so that solvers can find suitable solutions. Systematic strategies must also be developed to refine the programmers' specifications when they are too weak and return too many matches, and to relax the constraints in order to find partial matches when exact ones are not available. Techniques to support the composition of partial matches will also be necessary to scale the approach to larger and more diverse search queries. Last, infrastructure must be built and studies conducted to determine whether the proposed approach can be cost-effective in practice. Use of these techniques will change the way programmers operate, directly impacting their productivity by enabling them to truly leverage existing code in increasingly rich code repositories.