GHC Commentary: The Runtime System
GHC's runtime system is a slightly scary beast: 50,000 lines of C and C-- code, much of which seems at first glance to be completely obscure. What on earth does the RTS do? Here are the highlights:
It includes all the bits required to execute Haskell code that aren't compiled into the code itself. For example, the RTS contains the code that knows how to raise an exception when you call
error, code to allocate
Array#objects, and code to implement
It includes a sophisticated storage manager, including a multi-generational garbage collector with copying and compacting strategies.
It includes a user-space scheduler for Haskell threads, together with support for scheduling Haskell threads across multiple CPUs, and allowing Haskell threads to call foreign functions in separate OS threads.
There's a byte-code interpreter for GHCi, and a dynamic linker for loading up object code into a GHCi session.
Heap-profiling (of various kinds), time-profiling and code coverage of Haskell code are included.
Support for Software Transactional Memory.
Next, we try to make sense of how it all fits together.
Find your way around the code
Basics you should know about
- Storage: memory layout and garbage collection
- Haskell Execution: how Haskell code is executed
- The Scheduler: threads, multi-processor support, FFI
- Sanity Checking
- So how does foreign import "wrapper" work?
- GHCi support: the byte-code interpreter and dynamic linker
- Asynchronous exceptions
- Software Transactional Memory (STM)
- Weak Pointers and Finalizers
- How Signals are handled
- The IO Manager thread
- The HEAP_ALLOCED macro
- Memory ordering
Also check the list of cross-cutting concerns in Commentary.