CADOS: Configurability-Aware Development of Operating Systems

CADOS的目的是研究可扩展的方法和技术，以掌握系统软件的可变性。例如，Linux 4.2为特定于问题的适应提供了15000多个静态可配置特性，这些特性在技术上实现并分布在生成过程的几个级别（配置系统、构建系统、C预处理器、链接器等）。由此产生的可变性在实践中不再是可管理的。在当前的资助阶段，从系统软件开发人员的角度出发，这已经集中在（现有的）静态可变性及其跨多个级别和工件语言的实现上进行了检查。所开发的方法和工具显著地改进了高度可配置的、与系统相关的软件中的静态分析，以及在它们的实现中（跨级别）可变性错误的检测（“#ifdef bugs”）。从“开发者的角度”（可变性的提供者）来看，这是朝着全面掌握系统软件中的静态可变性迈出的决定性的一步。在接下来的资助阶段，重点将扩展到非静态可变性的形式，但特别是“用户视图”（可变性的专业用户）：为操作系统从成千上万的可配置特性中手动选择“正确”的特性已经是一项几乎不可能的任务。此外，对于特定的应用程序，还必须考虑到大量的其他系统软件（库、服务器进程、实用程序）。因此，在实践中应用的最大配置（标准分布）携带了许多不必要的和昂贵的“cruft”，而由于真正的替代功能（特殊指令的可用性，单核或多核，使用的中断控制器），仍然需要多种变体。最终的系统软件既“太大”（关于不需要的选项）又“太小”（关于需要但依赖于上下文的替代选项）。因此，下一个筹资阶段的研究课题是分析、实施和应用“半动态变异性”，作为在语言和编译器一级以及在操作系统的必要支持方面实现系统软件中灵活和依赖于上下文的可选和可选特性的一种手段。另一个研究课题是“自动全系统剪枝”，通过跨层边界分析整个系统软件堆栈，以便在很少的人工操作下自动配置和剪枝最终的系统软件。在研究中，我们将主要应用于操作系统构建（内核、链接器和加载器、监控）和编译器（语言、静态分析和代码生成）领域的建设性方法。本课题的研究对象是开源系统软件。

Purpose of CADOS is to research scalable methods and techniques to master the variability of (and in) system software. Linux 4.2, for example, offers more than fifteen thousand statically configurable features for problem-specific adaptation, which are technically implemented and distributed over several levels of the generation process (configuration system, build system, C preprocessor, linker, ...). The resulting variability is no longer manageable in practice.In the current funding phase, this has been examined with focus on (existing) static variability and its implementation across multiple levels and artifact languages from the viewpoint of system-software develpers. The devloped methods and tools significantly improve static analysis in highly configurable, system-related software, as well as the detection of (cross-level) variability errors in their implementation ("#ifdef bugs").From the "developer's point of view" (provider of variability), this is a decisive step towards the comprehensive mastery of static variability in system software.In the following funding phase, the focus shall be expanded to forms of non-static variability, but in particular to the "user view" (professional users of variability): Manually selecting the "right" features for a specific problem from tens of thousands of configurable features for the operating system is already a nearly inpossible undertaking. In addition, there is an extensive stack of other system software (libraries, server processes, utilities) that must be taken into account as well for the specific application.Hence, maximum configurations (standard distributions) are applied in practice that carry a lot of unneeded and costly "cruft", while there are still multiple variants required due to genuinely alternatvive features (availability of special instructions, single or multicore, interrupt controller used). The resulting system software is both “too big” (with regard to options that are not required) and “too small” (with regard to required but context-dependent alternatives).The subject of research for the following funding phase is therefore the analysis, implementation and application of "semi-dynamic variability" as a means of realizing flexible and context-dependent optional and alternative features in system software, both at the language and compiler level and with regard to the necessary support from the operating system.The other subject of research is "automatic system-wide pruning", through which the entire system software stack is to be analyzed across layer boundaries in order to automatically configure and trim the resulting system software with little manual effort.In the research we will apply mostly constructive methods in the domains of operating-system construction (kernel, linker and loader, monitoring) and compilers (language, static analysis and code generation). The subject of research is open-source system software.