Disable cse_global on create_lb_collision_rule for Sympy < 1.2

I have this weird problem with cse_global and the fluctuating LB. It works fine with SymPy 1.2 and produces wrong temperatures with SymPy 1.1.1. I have bisected it down to this change: https://github.com/sympy/sympy/pull/13221. My patch disables cse_global if it's enabled and displays a warning on the incriminating SymPy version. I used Mathematica to plug the subexpressions back into the main expressions and get the same thing with both SymPy versions, so I have no idea what the actual problem is. Thus I can't rule out that it only occurs with the fluctuating LB. Until we know for sure, it seems safest to just disable cse_global on that old SymPy version entirely.

added 1 commit

• 46dfc3f2 - Disable cse_global on create_lb_collision_rule for Sympy < 1.2

Compare with previous version

I think we need to investigate this in more detail, where the problem comes from. Your comment suggests that the sympy cse performs correctly, since inserting the terms back gives them same result.. is that correct?

Perhaps there is an error in the code generation backend that occurs only in certain cases, that the CSE produces

Your comment suggests that the sympy cse performs correctly, since inserting the terms back gives them same result.. is that correct?

That is correct.

Perhaps there is an error in the code generation backend that occurs only in certain cases, that the CSE produces

That sounds like a reasonable explanation. However, nothing about The expressions produced looks unusual, so I don‘t know why they work in one case and fail in the other.

Can you please take a look at this? lbmpy_tests/test_fluctuating_lb.py from !16 (merged) clearly shows the error.

The generated code contains

const int64_t xi_197 = 1.73205080756888;

which is obviously wrong.

This was caused by bug fixed in pystencils@721fdf45

when this fix is merged we can close this

mentioned in merge request !17 (merged)

closed

mentioned in issue pystencils#19 (closed)