Decimal import time

[GoogleCodeExporter]

Importing the decimal module significantly slows down gmpy startup. Can
this be avoided?

Original issue reported on code.google.com by fredrik....@gmail.com on 3 May 2009 at 10:13

[GoogleCodeExporter]

Is checking for an environment variable GMPY_NO_DECIMAL (or similar) an 
acceptable
solution? 

(I wonder if importing the decimal module is a valid approach if decimal is 
rewritten
in C?)

Original comment by casevh on 4 May 2009 at 6:02

• Added labels: Type-Enhancement
• Removed labels: Type-Defect

[GoogleCodeExporter]

Why is the code needed in the first place?

If I recall correctly, Decimal used to raise TypeError on unknown types but 
this has
been fixed to return NotImplemented in recent Python versions. It should be 
possible
to do everything in gmpy's coercion code, without monkeypatching decimal.py.

Original comment by fredrik....@gmail.com on 4 May 2009 at 6:11

[GoogleCodeExporter]

Good point. I'll see what happens when I remove it.

Original comment by casevh on 4 May 2009 at 6:25

[GoogleCodeExporter]

In the current trunk versions, I've removed the automatic import of Decimal. In
combination with the changes required for Python 3.x support, I ended up 
removing
support for coerce since it isn't supported with Python 3.x. This led to various
subtle changes with number conversions. I tried to mimic Python's behavior as 
much as
possible. In gmpy 1.04, mpz(1) + 1.2 returns mpz(2). I think the implicit 
conversion
of 1.2 to 1 is incorrect. With the new logic in trunk, this will generate a
TypeError. The alternative would be to return an mpf, but then I'd need to 
guess the
desired precision. I think conversions that can discard information should not 
be
done automatically.

Do you have any preferences?

Original comment by casevh on 8 Jun 2009 at 7:21

[GoogleCodeExporter]

I get a float:

>>> import gmpy
>>> gmpy.mpz(2) + 1.2
3.2000000000000002

I think there is some code that relies on this.

Original comment by fredrik....@gmail.com on 8 Jun 2009 at 9:41

[GoogleCodeExporter]

You're correct. I got confused with all the revisions I've been testing. The 
current 
behavior has quirks such as:

>>> gmpy.version()
'1.04'
>>> gmpy.mpz(10) == 10.0
False
>>> gmpy.mpz(10**400) + 1.2
inf
>>> gmpy.mpz(10) + decimal.Decimal('10.2')
mpz(20)

I probably got confused with the current state of Decimal conversion. How 
important 
is consistency between Decimal and float? I think the most consistent behavior 
would 
be either raise an error and force explicit conversion, or convert to mpf and 
guess 
at the precision. All this really does is consistently break both scenarios, 
though. ;-)

Original comment by casevh on 8 Jun 2009 at 11:57

[GoogleCodeExporter]

The automatic import and monkey-patching of Decimal has been removed in gmpy 
1.10.

Original comment by casevh on 5 Jul 2009 at 5:13

• Changed state: Fixed