HACKING.rst

node.py resolve.py

class method node.ident self.ident

HACKING

	matlab	fortran	python
Base-one indexing	yes	yes	no
Columns-first data layout	yes	yes	no
Auto-expanding arrays	yes	no *	yes
Update to create	yes	no *	yes
Assignment as copy	yes	yes	no

	matlab	fortran	python
Matrices everywhere	yes	no	no
Single subscript implies ravel	yes
Broadcast
Boolean indexing
Type and rank must be known in compile time	no	yes	no

	matlab	fortran	python
Garbage collection	yes	no *	yes
All uppercase	no	yes	no
Structs
Interpreted	yes	no	yes
Strings are arrays of chars	yes	no	yes

3. Auto-expanding arrays

Matlab arrays are auto-magically resized on out-of-bounds update. Deprecated, this feature is widely used in legacy code. Supporting this feature is hard both in python and in fortran.

In fortran, the pattern should be somehow (how exactly?) detected in compile-time. In python __setitem__ hides try-catch, with resize called inside catch. Is try-catch in fortran?

In numpy out-of-bounds assignment is an error. In smop, out-of-bounds assignment is supported for row and column matrices and their generalizations having shape

[1 1 ... N ... 1]

These arrays may be resized along their only non-singular dimension. For other matrices, new columns can be added to F_CONTIGUOUS arrays, and new rows can be added to C_CONTIGUOUS arrays.

Finally, scalar array of any dimension, having shape

[1 1 ... 1]

can be resized along any dimension.

4. Update to create

In matlab, arrays may be created by updating a non existent array, as in the example:

>>> clear a
>>> a(17)=42

This unique feature is not supported by smop, but can be worked around by inserting assignments into the original matlab code:

>>> a=[]
>>> a(17_=42

---