Its all there in SymPy, just need @sym/sym.m constructor work.

snan = sym(0)/0
snan == 1
islogical(snan == 1)

(inf seems ok, unit tests done to keep it that way)

This works at our end but fails in SymPy, issue with ImmutableMatrix().

Todo: find or post an upstream issue?

We have prod.m, should be similar.

see comments in the files

Needs work in subsref, subsasgn and subindex

Several ways to make 1x1 matrices which in sympy are not same as scalars:

1. inner product for example
2. grow an empty sym into a 1x1 sym
3. (probably) via slicing (but that doesn't work yet, Bug #14)

One way to fix this is on conversion back from python, replace them with scalars.

syms x
x(1) % fails
x(1) = sym(2) % works

also, add this test to subsref

x = sym([1 2; 3 4])
cos(x)

Functions of SymPy Matrix() don't do component evaluation. Can use .apply_func instead in the code generator.

I've overloading subrefs() and it works fine. But within my own @sym/ directory, I have to manually use subrefs() like this:

x(i)   % <-- gives error
% so workaround:
idx.type = '()';
idx.subs = {j};
subsref(x, idx);

The error is:

error: myfcn.m: A(I): index out of bounds; value 2 out of bound 1
error: called from:

It's calling the superclass subsref (I think). Which in my case, will report [1 1] or something wrong for my array.

FWIW, same thing in Matlab 2008:

??? Index exceeds matrix dimensions.

SMT gives an empty sym

I've gone back and forth on this:

1. Using SymPy's Matrix for arrays. Currently the master branch does this.

2. Use native arrays (only scalars go to SymPy). See octarray branch for this approach.

Anyway, make a decision and close this :)

need sympy codegen

e.g.,

syms x y real
f = 2*x
assumptions()
assumptions(f)

in matlab, you can do "syms f(x)". Currently we need "syms x f(x)".

mat_access.m and friends for example.

Matlab SMT gives a subfym:

>> syms f(t)
>> f
f(t) =
f(t)
>> g = 2*f
g(t) =
2*f(t)
>> h = cos(f)
h(t) =
cos(f(t))
>> whos
  Name      Size            Bytes  Class     Attributes

  f         1x1               112  symfun              
  g         1x1               112  symfun              
  h         1x1               112  symfun              
  t         1x1               112  sym                 

this is b/c of the .copy change I made to make dict's work...

should make each of x,y,z positive

check SMT, but I assume it should do elementwise diff

x = sym(1)
x(2) = 2
% error

Workaround: It can be escaped as %% in python_cmd() and as %%%% in evalpy.

This is due to gratuitous sprintf() when all we really want is to unescape \n.

same as just syms x really. But it should probably loop over all syms like assume...

Currently, sym('bar') calls S('bar') in SymPy. S() is smart. S('beta') gives the beta function. Perhaps we don't want that.

1. We might sometimes want something dumb, like Symbol('beta'). But then Symbol('1') is not One, its just a symbol, which is not good. So we'd have to work around for numbers.
2. Alternative? detect special inputs like 'gamma' and 'beta' and pass them to Symbol instead of S? Then we have a huge list of special cases to maintain. But surely better than the above? At least oo, nan, zoo would still work as expected. I think I'll do this for now.

FWIW, Matlab SMT does 1. It allows plain beta, gamma, etc: "syms beta gamma". They also also "syms sin". I.e., its complete user's responsibility. If she wants a symbol called "sin" or "limit" then so be it.

see taylor.m. Hard-coded for 8 terms for example.

something about numel? at least on 3.6.4

a = [1 x]'

gives

[1; conj(x)]

but

x'

does not give

conj(x)

Should be an easy fix as part of Bug #19.

This is a problem with the Octave parser. Its fixed in hg so should be ok in Octave 4.0.

A = [sym(0) 1; sym(2) 3];
B = [sym(0) 1; 2 3];

A works but B fails on Octave 3.6.4 and 3.8.1 with:

error: octave_base_value::map_value(): wrong type argument 'scalar'
error: octave_base_value::parent_class_name_list(): wrong type argument 'scalar'

It works in Matlab, maybe an Octave bug?

One way is probably with static methods, which i think needs classdef.

So wait for octave to get classdef?

python generates C "%a" formatting:

0.11 -> '0x1.c28f5c28f5c29p-4'

java might work, or write some custom code. In first instance, I'll use a string with many digits but it would be good to pass the bit-exact double precision value back from SymPy.

See the %! xtest in symvar.m if you want to try to fix this. I think the code for sorting is incorrect for capital letters. Should write some string sort function (maybe there are some?)

Or easier, do it in Python (see findsymbols.m, which sorts in lexiographic order).

I fear this will be tricky.

Some commands return matrices of bools. We want these to become "logical"s in Octave. Currently dealt with in an adhoc manner (e.g., see isprime).

Perhaps something more general python_header.py would be good?

Script should run them too

essentially works, but returned thing is not useful (or wasn't last time I looked).

Need to pass dict back from python, probably to a struct.

Currently, you can "hide" assumed variables inside e.g., cell-arrays

stderr does not seem to appear on the screen. Is it redirected to stdout? Possible to control it?