improve converting Latex to SymPy

[bhpayne]

SymPy is capable of converting Latex to SymPy. It works for simple Latex, but complicated expressions do not result in the desired SymPy representation.

[bhpayne]

SymPy converts a simple Latex expression to SymPy representation:

>>> import sympy
>>> sympy.__version__
'1.8'
>>> from sympy import *
>>> from sympy.parsing.latex import parse_latex
>>> sympy.srepr(parse_latex('T = 1/f'))
ANTLR runtime and generated code versions disagree: 4.9.2!=4.7.2
"Equality(Symbol('T'), Mul(Integer(1), Pow(Symbol('f'), Integer(-1))))"

Integration:

>>> sympy.srepr(parse_latex('f = \int_a^b x dx'))
ANTLR runtime and generated code versions disagree: 4.9.2!=4.7.2
"Equality(Symbol('f'), Integral(Symbol('x'), Tuple(Symbol('x'), Symbol('a'), Symbol('b'))))"

[bhpayne]

SymPy fails on this Latex

>>> import sympy
>>> sympy.__version__
'1.8'
>>> from sympy import *
>>> from sympy.parsing.latex import parse_latex
>>> sympy.srepr(parse_latex('\vec{\psi}(r, t)'))
antlr4.error.Errors.LexerNoViableAltException: LexerNoViableAltException(
sympy.parsing.latex.errors.LaTeXParsingError: I don't understand this

The desired SymPy representation would be something like

Function('psi')(Symbol('r'), Symbol('t')

One reason this example is difficult because the dimensionality of \vec{\psi} is undefined. If we impose the assumption of 3 dimensions, SymPy does have

>>> from sympy.vector import CoordSys3D, Del, curl, divergence, gradient

so it is feasible to define

>>> N = CoordSys3D('N')
>>> psi_vector = Symbol('a')*N.i+Symbol('b')*N.j + Symbol('c')*N.k

This seems to necessitate a multi-line SymPy operation to support a single (ill-defined) Latex expression.

[msgoff]

The grammar has more than a dozen logical errors.

It would be ideal to reuse the non-recursive terminals, and to rewrite the recursive rules, with the primary goal of improving correctness while not breaking rules that people depend on.

ideas on how to proceed would be welcome.

I think the following would be helpful.
Having one or more people to communicate with that would care about the output.
Generating useful test cases that are required for updates to the grammar.