Imports in Python-Flint on Windows

[BrianGladman]

HI, My apologies for posting an end-user question here but I couldn't find an alternative place to pose it.

I have a python program that uses rationals and I am trying to test it with various libraries including python-flint.

My code works when I use the native Python import: "from fractions import Fraction as RF"

It also works when using the GMPY2 module import: "from gmpy2 import mpq as RF"

But using: "from flint import fmpq as RF" fails with the error :

builtins.TypeError: cannot create fmpq from object of type <class 'flint.types.fmpq.fmpq'>

Using: "from flint import *" followed by "RF = fmpq" fails with the same error

Using: "from flint import *" followed by "RF = flint.fmpq" fails with builtins.NameError: name 'flint' is not defined

So I have not found a way of importing fmpq from Python-Flint :-(

How should I be doing this?

with my regards,

   Brian Gladman

[oscarbenjamin]

The code as shown should work:

In [2]: from flint import fmpq as RF

In [3]: RF(2,3)
Out[3]: 2/3

I think the only way to see the error message you report is if passing two arguments of type fmpq to fmpq e.g.:

In [4]: RF(RF(1), RF(2))
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[4], line 1
----> 1 RF(RF(1), RF(2))

File flint/types/fmpq.pyx:97, in flint.types.fmpq.fmpq.__init__()

TypeError: cannot create fmpq from object of type <class 'flint.types.fmpq.fmpq'>

I think somewhere you are calling RF(a, b) where a and b are rational numbers rather than integers which should really be written as a / b. It seems that both Fraction and mpz allow that but fmpz does not. The valid constructors are:

fmpq()          # zero
fmpq(int)       # convert an integer to rational
fmpq(int, int)  # rational from pair of integers
fmpq(fmpq)      # rational from rational
fmpq(str)       # rational from string

[BrianGladman]

Hi Oscar, Thanks for your rapid response and your accurate diagnosis as I am indeed using a constructor that will at some point try to use the constructor fmpq(fmpq, fmpq).

Is this not a sensible constructor to implement? It seems to me a bit odd that two fmpq cannot make an fmpq as it is a somewhat awkward limitation to work around.

Should I turn this comment into a feature request?

[oscarbenjamin]

It can be a feature request. The fact that similar libraries provide this means that it is useful for compatibility if nothing else.

Is this not a sensible constructor to implement?

It is redundant because it is just dividing two fmpq:

In [30]: from gmpy2 import mpq as RF

In [33]: RF(RF(2,3), RF(5,7))
Out[33]: mpq(14,15)

In [34]: RF(2,3) / RF(5,7)
Out[34]: mpq(14,15)

The constructor is for turning things that are not fmpq into fmpq. If you already have two fmpq then you can just divide them.

I would tend to think that in my code it should be clear where I have fmpq vs fmpz because they behave differently and represent different domains. If I already know that have fmpq then I can just divide them which I think expresses more clearly what is happening: a / b rather than fmpq(a, b).