Question

All my unit tests succeed running in Python 2.6.5; one fails when I run through Python 2.7.3. The code being tested is complex and involves lots of working in floats and converting to Decimal along the way, by converting to str first as was needed in Python 2.6.

Before I start digging, I was wondering if I could be a bit lazy and see if someone has seen this before and has suggestions on what to search for. Here’s the result of the test run:

======================================================================
FAIL: test_hor_tpost_winsize_inside_mm (__main__.Test_ShutterCalculator)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "test_ShutterCalculator.py", line 506, in test_hor_tpost_winsize_inside_mm
    self.assertEqual(o.net_width_closing_tolerance, Decimal("6.4"))
AssertionError: Decimal('6.3') != Decimal('6.4')

----------------------------------------------------------------------

Here’s the unit test code for test_hor_tpost_winsize_inside_mm():

490     def test_hor_tpost_winsize_inside_mm(self):
491         """
492         Same as above but test mm
493         """
494         o = self.opening
495         o.unit_of_measure = "millimeters"
496         o.formula_mode = "winsize"
497         o.mount = "inside"
498         o.given_width = Decimal("1117.6")
499         o.given_height = Decimal("2365.4")
500         o.louver_spacing = Decimal("101.6")
501         self.make4SidedFrame("9613", '9613: 2-1/2" Face Deco Z', Decimal("63.5"), Decimal("19.1"))
502         so1 = o.subopenings[(0,0)]
503         so1.fold_left = 1
504         so1.fold_right = 1
505         self.calc()
506         self.assertEqual(o.net_width_closing_tolerance, Decimal("6.4"))
507         self.assertEqual(o.net_height_closing_tolerance, Decimal("6.4"))
508         self.assertEqual(o.horizontal_shim, Decimal(".125"))  # in inches
509         self.assertEqual(o.vertical_shim, Decimal(".125"))  # in inches
510         self.assertEqual(o.width, Decimal("1069.8"))  ## 1070 converted directly from inches
511         self.assertEqual(o.height, Decimal("2317.6")) ## 2317.8 converted directy from inches
512         tpost = o.add_hor_tpost()
513         so2 = o.subopenings[(0,1)]
514         so2.fold_left = 1
515         so2.fold_right = 1
516         self.calc()
517         #self.cs()
518         self.assertEqual(o.net_width_closing_tolerance, Decimal("6.4"))
519         self.assertEqual(o.net_height_closing_tolerance, Decimal("12.7"))
520         self.assertEqual(o.horizontal_shim, Decimal(".125"))  # in inches
521         self.assertEqual(o.vertical_shim, Decimal(".125"))  # in inches
522         self.assertEqual(o.width, Decimal("1069.8"))  ## Rick had 42 but agreed that mine is right
523         self.assertEqual(o.height, Decimal("2311.3"))
524         self.assertEqual(so1.width, Decimal("1069.8"))
525         self.assertEqual(so2.width, Decimal("1069.8"))
526         self.assertEqual(so1.height, Decimal("1139.7"))  ## Rick had 44.8125 but agreed mine is right
527         self.assertEqual(so2.height, Decimal("1139.7"))
528         self.assertEqual(tpost.center_pos, Decimal("1182.7"))
529         top_panel_section = so1.panels[0].sections[(0,0)]
530         bottom_panel_section = so2.panels[0].sections[(0,0)]
531         self.assertEqual(top_panel_section.louver_count, 9)
532         self.assertEqual(bottom_panel_section.louver_count, 9)
533         self.assertEqual(top_panel_section.top_rail.width, Decimal("112.6"))  ## Rick had 4.40625, but given the changes to net
534         self.assertEqual(bottom_panel_section.bottom_rail.width, Decimal("112.7"))
535         self.assertEqual(top_panel_section.bottom_rail.width, Decimal("112.7"))
536         self.assertEqual(bottom_panel_section.top_rail.width, Decimal("112.6"))

Any hint on what to search for in my code to find the source of the discrepancy?

Answer 1

Python 2.7 introduced changes to the Decimal class and float type to help improve accuracy when converting from strings. This could be the source of the change.

Conversions between floating-point numbers and strings are now correctly rounded on most platforms. These conversions occur in many different places: str() on floats and complex numbers; the float and complex constructors; numeric formatting; serializing and deserializing floats and complex numbers using the marshal, pickle and json modules; parsing of float and imaginary literals in Python code; and Decimal-to-float conversion.

You can see the change details here, under “Other language changes”