[Bug]: Inaccuracy introduced when non-zero fractional bits

[cameronpoe]

Summary

In the code I am trying to run, I am working with numbers that are over 64 bit with no fractional bits. At some point, I need to bitshift down and store in a new Fxp variable that has fractional bits. However, it seems that assigning an Fxp variable with non-zero fractional bits any whole number that is larger than 64 bits causes inaccuracy. I assume this is because something under the hood is converting to np.float64 or np.int64. However, this is not expected behavior.

Minimal reproducible example

a = Fxp(None, signed=True, n_int=63, n_frac=0, round='around', saturate='overflow')
a.set_val(2**63 - 1)

b = Fxp(None, signed=True, n_int=65, n_frac=0, round='around', saturate='overflow')
b.set_val(a + a + a)
print(a)
print(b)
print(a.info(verbose=3))
print(b.info(verbose=3))

c = Fxp(None, signed=True, n_int=65, n_frac=2, round='around', saturate='overflow')
c.set_val(a + a + a)
print(c)
print(c.info(verbose=3))

Actual behavior

9223372036854775807
27670116110564327421
dtype = fxp-s64/0
Value = 9223372036854775807
inaccuracy = True
extended_prec = True

Signed		=	True
Word bits	=	64
Fract bits	=	0
Int bits	=	63
Val data type	=	<class 'int'>

Upper		=	9.223372036854776e+18
Lower		=	-9.223372036854776e+18
Precision	=	1.0
Overflow	=	saturate
Rounding	=	trunc
Shifting	=	expand

None
dtype = fxp-s66/0
Value = 27670116110564327421
inaccuracy = True
extended_prec = True

Signed		=	True
Word bits	=	66
Fract bits	=	0
Int bits	=	65
Val data type	=	<class 'int'>

Upper		=	3.6893488147419103e+19
Lower		=	-3.6893488147419103e+19
Precision	=	1.0
Overflow	=	saturate
Rounding	=	trunc
Shifting	=	expand

None
2.7670116110564327e+19
dtype = fxp-s68/2
Value = 2.7670116110564327e+19
inaccuracy = True
extended_prec = True

Signed		=	True
Word bits	=	68
Fract bits	=	2
Int bits	=	65
Val data type	=	<class 'float'>

Upper		=	3.6893488147419103e+19
Lower		=	-3.6893488147419103e+19
Precision	=	0.25
Overflow	=	saturate
Rounding	=	trunc
Shifting	=	expand

None

Expected behavior

I would expect the b and c to be the same exact value, except that c has two fractional bits that are 0.

fxpmath version

0.4.10

Python version

3.13.2

NumPy version

2.3.1

Operating system

Ubuntu 26.04

Additional context

No response