There is a long way to go between the current implementation and a final robust one. The final version should be able to emulate the vast majority of MATLAB/Octave language syntax, and should also try to emulate some of the basic library functions thereof. Notice that some "standard" features of MATLAB/Octave won't be available here, at least not immediately: Plots and graphs, user interfaces, toolboxes, etc.
Version 1.0 will support most of the basic language syntax, and hopefully many of the core library functions. Here, abbreviated and horribly inaccurate, is a general roadmap for this project:
Version 0.1: (current) We have basic support for single- value subroutines, basic control structures, basic scalar variables. Also some ideosyncratic features like using ';' to suppress printing line results, using '!' to make a system call, etc. Implemented basic functions like 'disp', 'error', 'exit', and 'feval', all for scalar arguments only.
Version 0.2: Implement matrices, and improved support for functions with multiple return values.
Version 0.3: Write some of the standard matrix manipulation functions, or basic versions thereof. Transpose, determinant, size, addition, multiplication, etc.
Version 0.4: Implement function/file look-up. Searching for function "foo", look in file "foo.m", etc.
Version 0.5: Implement cell arrays
Version 0.6: Write more-advanced function features: nargin, nargout, varargin, varargout, return. Implement matrix- ready functions that iterate over matrix values
Version 0.7: Implement matrix and cell array slicing: 'x(1:)', and similar. implement matrix and cell array concatination '[[1 2 3], [4 5 6]] = [1 2 3 4 5 6]'. Treat all scalars as matrices.
Version 0.8: Implement basic MATLAB classes and data structures. Implement a dedicated matrix PMC (probably in C) implement sparse matrices (may be part of the same pmc).
Version 0.9: Implement most of the standard library of functions, including error-stubs for common functions which cannot be implemented (graphics, etc). Core builtins should be written in PIR, and the rest should be written in M.
Version 1.0: All the basic stuff should be working, including a proper test suite. Enough of the language should be implemented in order to run the Octave test suite (or portions of it that we haven't explicitly excluded). Matrixy won't have all the features of MATLAB or Octave, but the features it does have should match them syntactically. Everything should be documented
The roadmap above should produce a 1.0 which is a stable clone of a core subset of MATLAB/Octave. From that base, there are a large number of standard features (extensions/libraries, larger subsets of the language) that could be implemented. Plus, there are a number of non-standard wishlist features that could be added to extend the language. Such non-standard features might be made available through a command-line switch, an in-code pragma or in a separate executable. Here is a list of such features, in no particular order:
=item* Better, more robust support for OO, classes, namespaces.
Notice that OO was added to MATLAB in R2008A. This wishlist item means better support for this including some of the more advanced features that Parrot enables.
=item* Better Multithreading
Notice that MATLAB supplies a Parallel Computing Toolbox which gives the user some explicit support for multithreading and support for multicore computing. Matrixy may attempt to reproduce this toolbox or may attempt to add a more generic threading system.
=item* disambiguous matrix/function syntax.
Should be able to use '' for matrix indices, use '()' for function arguments, if desired. This will help ease human readability in some cases, and is more natural for some programmers.
=item* Allow inlining code from other languages.
Matrixy will likely contain subroutines very early on to compile code strings from other programming languages. An inlineable interface will be more robust and natural.