Full-codegen

Full-codegen
Name	Full-codegen
Developer	Mozilla Corporation
Released	2011
Latest release	2016
Programming language	C++
Platform	x86, ARM
License	MPL

Full-codegen

Full-codegen is a now-deprecated code generation backend originally developed within the Mozilla Corporation project ecosystem. It served as an emitting layer for the SpiderMonkey JavaScript engine, producing machine code for x86, x86-64, and ARM architectures and interacting with runtime components such as the JIT pipeline, the garbage collection subsystem, and the runtime linker on a variety of operating systems including Linux, Windows, and macOS. As part of a broader effort inside projects like Firefox and experimental work influenced by compilers from Google and Apple, Full-codegen prioritized simplicity and portability at the cost of some runtime performance characteristics.

Overview

Full-codegen functioned as a straightforward backend that translated an intermediate representation produced by earlier tiers of the SpiderMonkey compiler into platform-specific native instructions. It occupied a position alongside other components such as the JIT tier, the IonMonkey optimizing compiler, and the Baseline compiler, coordinating with systems like the XPCOM interface, the NSS crypto library, and the Gecko rendering engine to deliver executable code to the browser runtime. The design emphasized predictable compilation latency for web workloads similar to those targeted by browsers like Chrome and Safari, and it interfaced with tooling and debugging facilities used by projects such as GDB and LLDB.

History and Development

Development of Full-codegen occurred during a period of intensive competition among browser vendors, with influence from research at institutions like MIT, Stanford University, and UC Berkeley on dynamic language implementation. Early milestones were recorded in coordination with releases of Firefox 4 and later versions, during which Full-codegen complemented efforts by teams at Mozilla Foundation and partner contributors including individuals formerly associated with Netscape and researchers who published at conferences such as PLDI, OOPSLA, and CGO. The push for more advanced optimizers led to projects like IonMonkey and later Warp, resulting in Full-codegen being gradually superseded. Contributions came from engineers who previously worked on systems such as V8 at Google and JavaScriptCore at Apple, reflecting cross-pollination across projects like Chromium and WebKit.

Architecture and Implementation

Full-codegen’s architecture comprised a pipeline that accepted a lowered representation from the frontend parser and bytecode engine, mapping constructs to machine-code sequences using templates tuned for targets including ARMv7, ARM64, Intel Pentium, and AMD64 processors. It integrated with the SpiderMonkey runtime components such as the JSRuntime, the JSContext, and the GC marker, emitting prologues and epilogues compatible with calling conventions used by libc on platforms like FreeBSD and OpenBSD. Implementation choices reflected trade-offs visible in compilers like LLVM and projects such as GCC: Full-codegen favored compact emitter code and maintainability rather than aggressive global register allocation strategies pioneered in systems like Ollvm or research prototypes at Carnegie Mellon University. The backend provided trampolines, inline caches analogous to techniques reported by researchers at Sun Microsystems, and basic stack frame management interoperable with debuggers such as Valgrind and profilers associated with perf.

Performance and Optimization

Full-codegen’s performance profile emphasized low compile overhead and predictable latency to benefit interactive environments exemplified by browsers such as Firefox, Chrome, and Edge. Compared to the highly optimizing IonMonkey and engines like V8’s Turbofan, Full-codegen generated more conservative instruction sequences, similar to earlier JITs described in literature from University of Cambridge and ETH Zurich. Optimization strategies included peephole optimizations, simple constant folding, and inline-cache management akin to techniques used in Smalltalk virtual machines and research from Brown University. Where microbenchmarks used by organizations such as Mozilla and third-party groups highlighted throughput differences, Full-codegen’s strengths were in startup time and memory footprint, with trade-offs in hot-path throughput when contrasted with advanced trace or SSA-based optimizers developed at Microsoft Research and Yahoo!.

Use Cases and Adoption

Full-codegen’s primary use case was within the Firefox browser as a reliable code-emission stage for everyday web pages, interacting with platform integrations such as WebExtensions, NPAPI legacy interfaces, and network stacks leveraging NSS and Necko. It saw adoption in contexts requiring fast cold-start behavior, similar to deployment concerns faced by teams at Amazon when optimizing client-side workloads, and it influenced teaching examples at institutions like University of Illinois and University of Washington for courses on virtual machines and dynamic compilation. External projects and forks within the open-source community, including mirrors hosted by organizations such as GitHub and collaborations involving contributors from Red Hat and Canonical, referenced Full-codegen while developing alternative backends or studying SpiderMonkey internals.

Limitations and Criticisms

Critics pointed to Full-codegen’s conservative optimization approach as limiting peak performance compared with optimizing compilers like IonMonkey or Turbofan, and researchers citing benchmarks from groups such as SPEC and JetStream noted disparities on compute-heavy workloads. The backend’s maintenance burden was also discussed in the context of engineering resource allocation at organizations like Mozilla Foundation and led to deprecation in favor of more unified pipelines inspired by work at Google and Apple. Security analysts referencing incidents involving components in complex runtimes such as OpenSSL and libpng emphasized the need for rigorous testing and fuzzing, and Full-codegen’s simpler design was both praised for reduced surface area and critiqued for lacking advanced mitigations present in later systems examined in publications from CERT and USENIX.

Category:Compilers