Julia Evans

How are Ruby's headers different from Python's headers?

This is another research-y post, most things in here I learned in the last 24 hours so likely some of it is wrong.

Today and yesterday I’ve been trying to figure out how the public header files Ruby exposes are different from the header files Python exposes (how are their public C APIs different?). You can get these headers in the libruby-dev / libpython-dev Debian packages. These header files live in /usr/include/{ruby-2.3.0|python-VERSION} on my laptop, and they’re one of the interfaces that Python/Ruby provide that I don’t usually think about that much. These header files are what you use when you’re writing a C extension for the language (like numpy) or embedding the language.

My stake in this is that I’m trying to do for Ruby essentially what pyflame does for Python, so understanding how Ruby is different from Python is very useful to me. pyflame uses Python’s public headers. Can my profiler use the corresponding Ruby headers? I don’t think so!

A reason these are important is that there’s an expectation that they be somewhat stable – obviously they change from version to version sometimes, but they need to be more stable than Python/Ruby’s internal header files. The structs I was talking about yesterday are all internal Ruby VM structs, so they change a lot.

Python is embeddable

One fact that I learned yesterday is – you can embed Python in your C programs! Here are the docs on embedding Python. Some examples of programs that embed Python include Sublime Text, Maya, Blender, Inkscape, Nuke – a lot of graphics / 3D modelling programs. There are more listed at on Python’s Wikipedia page.

I also found this argument/rant from Glyph that you shouldn’t embed Python interesting reading.

Ruby is embeddable

There isn’t any documentation on http://ruby-lang.org (that I can find) about embedding Ruby. But when I said on twitter that ruby wasn’t embeddable, Matz replied and said “CRuby is also embeddable. But mruby has better API for embedding.” Obviously he’s right that CRuby is embeddable, so let’s find out what’s up!

First – https://github.com/mruby/mruby is a lightweight implementation of Ruby intended for embedding. The README says that the syntax is Ruby 1.9 compatible. Neat!

But what about CRuby being embeddable? Let’s investigate that. The first point of documentation is doc/extension.rdoc in the Ruby repo (previously called README.ext). The most useful docs I found was the Pragmatic Programmer chapter “Extending Ruby”.

mruby seems to be more popular for embedding than CRuby but at least one popular program embeds CRuby/MRI: RPG Maker! (thanks to Florian for pointing this out)

Cool example of embedding Ruby: filtering syscalls!

Quick tangent: this slide deck on hijacking syscalls with Ruby from Franck Verrot is really interesting! Basically it describes a way to make sure that your programs are only running ‘allowed’ system calls. Here’s how it works as far as I can tell:

Create a .so library that uses embedded ruby to load a ‘policy.rb’ file that dynamically filters / logs system calls Add that library to your LD_LIBRARY_PATH for the programs you want to check

The slide deck says that this was reasonably efficient.

Anyway, this is not about cool ways to embed Ruby. Let’s move on.

Ruby bindings: Everything is a VALUE

One surprising/interesting thing to me about the Ruby headers – the type returned by almost every function is VALUE. The equivalent to that for Python seems to be PyObject but there are more types than PyObject in there. And PyObject has a struct definition, whereas VALUE is an opaque pointer (typedef uintptr_t VALUE;).

To get an idea of what “everything returns VALUE” looks like in practice, here are the functions you can use to define classes in Ruby:

VALUE  rb_define_class(char *name, VALUE superclass")
   Defines a new class at the top level with the given name and superclass (for class Object, use rb_cObject). 
VALUE  rb_define_module(char *name")
   Defines a new module at the top level with the given name. 
VALUE  rb_define_class_under(VALUE under, char *name, VALUE superclass")
   Defines a nested class under the class or module under. 
VALUE  rb_define_module_under(VALUE under, char *name")
   Defines a nested module under the class or module under. 
void  rb_include_module(VALUE parent, VALUE module")
   Includes the given module into the class or module parent. 
void  rb_extend_object(VALUE obj, VALUE module")
   Extends obj with module. 
VALUE  rb_require(const char *name")
   Equivalent to ``require name.'' Returns Qtrue or Qfalse. 

Structs Ruby and Python expose in their bindings

Let’s talk about structs! In the public header files Ruby and Python export (which live in the libruby-dev / libpython-dev packages in Debian), there are struct definitions. How are those different? I found it useful to just list all the structs they export. Here’s the list!

Ruby 2.3.0: RArray, RBasic, RClass, RComplex, RData, RFile, RMatch, RObject, RRegexp, RString, RStruct, RTyped.

Python 3.5: PyAddrPair, PyASCIIObject, PyAsyncMethods, PyBaseExceptionObject, PyBufferProcs, PyByteArrayObject, PyBytesObject, PyCellObject, PyCFunctionObject, PyCodeObject, PyCompactUnicodeObject, PyCompilerFlags, PyComplexObject, PyCoroObject, PyCursesWindowObject, PyDateTime, PyDescrObject, PyDictObject, PyFloatObject, PyFrameObject, PyFunctionObject, PyFutureFeatures, PyGC, PyGenObject, PyGetSetDef, PyGetSetDescrObject, PyHash, PyHeapTypeObject, PyImportErrorObject, PyInstanceMethodObject, PyInterpreterState, PyListObject, PyLockStatus, PyMappingMethods, PyMemAllocatorDomain, PyMemAllocatorEx, PyMemberDef, PyMemberDescrObject, PyMemoryViewObject, PyMethodDescrObject, PyMethodObject, PyModuleDef, PyNumberMethods, PyObject, PyObjectArenaAllocator, PyOSErrorObject, PySequenceMethods, PySetObject, PySliceObject, PySTEntryObject, PyStopIterationObject, PyStructSequence, PySyntaxErrorObject, PySystemExitObject, PyThreadState, PyTracebackObject, PyTryBlock, PyTupleObject, PyType, PyTypeObject, PyUnicodeErrorObject, PyUnicodeObject, PyVarObject, PyWrapperDescrObject

Python gives you a lot more structs than Ruby, and those structs in general I think have more fields. The structs I want access to are like PyThreadState, and they’re not in Ruby’s public header files.

What interfaces does Ruby’s C API provide for getting stack traces?

An API for getting stack frames was added to Ruby in commit 774bff0adb on October 7, 2013 – the main function here seems to be rb_profile_frames. This is a function though so to call it you need to be in the same Ruby process, not in a separate process.

I went to look at the stackprof profiler and its initial commit is from October 10, 2013 – 3 days after the new profiler feature was released. Makes sense!

Here’s what stackprof’s initial release notes say:

samples are taken using a combination of two new C-APIs in ruby 2.1:

  • signal handlers enqueue a sampling job using rb_postponed_job_register_one. this ensures callstack samples can be taken safely, in case the VM is garbage collecting or in some other inconsistent state during the interruption.
  • stack frames are collected via rb_profile_frames, which provides low-overhead C-API access to the VM’s call stack. no object allocations occur in this path, allowing stackprof to collect callstacks for in allocation mode.

that’s all for now

There are probably more important differences between Ruby’s header files and Python’s header files that I don’t know about yet. That’s enough blogging for today though!