Oscar Benjamin

I thought it was used in more places like `__truediv__` but it seems not. Only initialising at first use seems reasonable.

> Actually `minpoly` doesn't even use `PurePoly` ``` In [8]: minpoly(sqrt(2), polys=True) Out[8]: PurePoly(_x**2 - 2, _x, domain='QQ') In [10]: type(minpoly(sqrt(2), polys=True).gens[0]) Out[10]: sympy.core.symbol.Dummy ``` It uses both PurePoly and...

@smichr do you have the repr for that matrix? It looks to me like something that would be solved efficiently by linsolve already although the complexity depends on how complicated...

It should be possible to reduce this efficiently with `linsolve`. Unfortunately if you select a subset of equations that is linear then `linsolve` gives the empty set: ```python In [26]:...

This also fixes a performance problem where match calls `as_dummy` which calls replace which calls match which calls as_dummy again. I don't even know how that avoids an infinite loop...

This is ready for review.

I've changed the PR title and edited the release note to explain more precisely what is changed.

Thanks

See also: https://packaging.python.org/en/latest/discussions/setup-py-deprecated/ Note that although that document says that setup.py is fine to use as a configuration for setuptools it is actually not recommended to configure setuptools that way...

The hatch docs have some discussion about differences between hatch and setuptools although only the section about the build backend is relevant here: https://hatch.pypa.io/1.12/why/