Matz's Ruby Implementation or MRI refers to the original Ruby interpreter by Yukihiro Matsumoto, the inventor of Ruby. Matz and contributors wrote the interpreter in C language; MRI is also known as C Ruby or CRuby, by analogy with CPython.
The term "MRI" excludes other Ruby engines (such as JRuby or Rubinius). For example, one can say that Ruby has Array#sort method, and MRI uses quicksort; other Ruby engines might use different sorting algorithm.
When code works with Ruby 1.8.7, this can be MRI 1.8.7, or anything else that implements the same language.
- MRI is the reference implementation, and the first implementation of every new Ruby version.
- MRI can fork, though only for Unix clones (not for Windows).
- MRI can save and restore continuations, though the implementation is slow because it copies the call stack.
- MRI has the Global VM Lock, alias Giant VM Lock or GVL. A thread, to run Ruby code, must hold this exclusive lock. Only one thread can hold the GVL; therefore, multiple threads can use only one CPU. (Contrast JRuby, where multiple threads can use multiple CPUs.)
- MRI 1.8 is slow. Programs for Ruby 1.8 often run faster in JRuby or Rubinius.
- an interpreter,
- a mark-and-sweep garbage collector, and
- the core and standard libraries.
Most of the core library is in C. The standard library is a mix of Ruby code and C extensions.
MRI 1.8 has a somewhat slow interpreter. MRI 1.9 has a new interpreter called Yet Another Ruby VM (YARV); it translates Ruby source code to an internal bytecode, then interprets the bytecode. Ruby code can run a few times faster in MRI 1.9 than in MRI 1.8. Ruby code remains slower than C code. For example, MRI 1.9.3 changes its 'date' package from Ruby code to a C extension; this gives better performance.