[Ironruby-core] Code Review: ClrPrimitives2

Tomas Matousek Tomas.Matousek at microsoft.com
Tue Mar 24 16:45:53 EDT 2009


From: Curt Hagenlocher
Sent: Tuesday, March 24, 2009 12:49 PM
To: Tomas Matousek; IronRuby External Code Reviewers
Cc: ironruby-core at rubyforge.org
Subject: RE: Code Review: ClrPrimitives2

Looks good.  The changes for the StringFormatter sites are much needed!

The test PythonInterop5 doesn't look complete -- there's no assertion.

Will add assertion.

It seems odd that CharOps would include Enumerable.  Is that because a char looks like a string that can never have a length != 1?

You can enumerate a single element :). Yes, it's for consistency.

You don't want to use Int32ToObject on ClrInteger.Narrow?

I'm not sure if it is needed. How often will the result be in the small range of numbers for which we cache objects?

There's a minor typo in the comment for MultiDimensionalClrArrayOps.


In RubyClass.GetDeclaredClrMethods, what if the member you're looking for actually has a name that ends in "*" ? :)

Then you should file a bug on Reflection ;-)


From: Tomas Matousek
Sent: Tuesday, March 24, 2009 11:33 AM
To: IronRuby External Code Reviewers
Cc: ironruby-core at rubyforge.org
Subject: Code Review: ClrPrimitives2

  tfpt review "/shelveset:ClrPrimitives2;REDMOND\tomat"

Python, DLR outer ring changes:

Hoists FloorDivision implementation to MathUtils - Ruby uses the very same operation.

Ruby changes:

A couple of breaking changes:

-          Removes ClrString constant. CLR string should be referred to as System::String.

-          Removes IronRuby library. "require 'IronRuby'" is no longer needed, IronRuby module is now a built-in module.

Implements integer/float operations for all CLR primitive numeric types (byte, sbyte, short, ushort, uint, long, ulong, float).
Those integer types that fit in 32-bit signed integer (Fixnum) are widened to Fixnum for all operations. The other integer types are widened to BigInteger. Float is converted to double.
The operations on these types don't narrow their results even if the values fit. For example, (System::Byte.new(1) + 1).class == Fixnum, not System::Byte. This might not be optimal for some scenarios. If we find it important we'll fix this in future.

The implementation takes the operations from FixnumOps, BignumOps and FloatOps and moves them to new modules IronRuby::Clr::Integer, IronRuby::Clr::BigInteger and IronRuby::Clr::Float respectively. These are mixed back into the numeric types. Some methods need to be specialized for each types, so we generated C# code for them. Ruby 1.9 script ClrInteger.Generator.rb produces ClrInteger.Generated.cs.

Implements System::Char and System::String methods so that they behave like an immutable UTF-8 encoded string (of size 1 character in the case of System::Char). Many methods in MutableStringOps can share implementation with CLR string. There is still a lot of work to be done here, especially to support encodings. So for now, to make CLR strings work like Ruby frozen strings to a large extent, I've implemented method-missing dispatch that forwards to a MutableString.

Adds IronRuby::Clr::MultiDimensionalArray that is mixed into all multi-dimensional arrays. This mixin needs to implement those IList methods that only work with vectors.

Enables checked arithmetic operations in debug builds. Fixes underflow/overflow bugs in integer operations and adds tests for them.

Adds C# code generator script - used for generating ClrInteger classes.
Groups all StringFormatter sites into a single site local storage class.
Fixes [#20265] library classes nested in an extension class have wrong name

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://rubyforge.org/pipermail/ironruby-core/attachments/20090324/955a0cbd/attachment-0001.html>

More information about the Ironruby-core mailing list