Details
-
Type:
Bug
-
Status: Closed
-
Priority:
Minor
-
Resolution: Fixed
-
Affects Version/s: 2.0.4, 2.1.1, 2.2
-
Component/s: Questions
-
Labels:
-
Testing Instructions:
-
Affected Branches:MOODLE_20_STABLE, MOODLE_21_STABLE, MOODLE_22_STABLE
-
Fixed Branches:MOODLE_20_STABLE, MOODLE_21_STABLE
-
Pull from Repository:
-
Pull Master Branch:
-
Pull Master Diff URL:
-
Rank:17885
Description
When using the significant figures calculation mode, an edge case is not properly handled in the function 'qtype_calculated_calculate_answer', causing the calculated answer to be incorrect.
The function works to ensure the answer is expressed as a string of the format "0.[1-9][0-9]*", but subsequently calls the round() function. In some cases, round will round up to 1.0- which is no longer in the assumed format. This causes an answer to be produced which is exactly 10x larger than the correct answer.
A special case should be added to compensate, as in the attached patch.
Activity
Kim,
Have you a numerical example that we could use to test.
However these ronding effects are also related to the PHP precision settings.
Try creating the following Simple Calculated Question:
All defaults, except:
Name: <whatever>
Answer Formula:
Answer Grade: 100%
Format: Significant Figures
Wildcard range for {x}
: 1.0-1.1
Decimal places: 1
Then, generate any large amount of wildcards. I chose 30. I receive the following error on any wildcard generated with the value 1.0:
"ERROR Correct answer : 100e-13 outside limits of true value 1.0E-12"
The primary problem is a rounding issue- which is more common than you'd think, given the way floating point numbers are stored internally.
Here's the problem described and demonstrated in a simple PHP script:
http://labs.ktemkin.com/moodle/bugs/rounding.phps
You can execute it on my server here:
http://labs.ktemkin.com/moodle/bugs/rounding.php
And with the proposed patch:
http://labs.ktemkin.com/moodle/bugs/rounding_patched.phps
http://labs.ktemkin.com/moodle/bugs/rounding_patched.php
Let me know if you have any questions.
Thanks Kim,
Things are ever more complex as PHP being an interpreted language that can tolerate such an expression as
{x}E-12.look at the following case
{x}
E-12 1.0E-12 = 10000e-13
Min: 9.8999999999999E-13---Max: 1.01E-12
Erreur de réponse correcte : 10000e-13 en dehors des limites de la valeur réelle 1.0E-12
This was from UQAM (my university) 1,9 site with calculated question.
This part of the code can be trace back to older version.
How do I got 10000e-13.
This was by setting the answer with 4 significant figures.
As long as the x is 1.0 (i.e. 1.000 ) all the expression is not correctly handled.
If the value is different as 1.00001 things are OK.
The only computer I have access (holidays...) is an Ipad .
Could you test if your patch applied to moodle code also corrected this bad rounding effect.
As a reminder, I also noticed that the left unit option is not handled correctly.
if (1 == $answerlength)
{ $calculated->answer = $sign.$answer.$exponent.$unit; } else {
// Attach additional zeros at the end of $answer,
Pierre, I think Kyle's fix is good, so submitting it for integration.
You are right that unit handling is wrong here, but I think we should handle that as a separate issue.
By the way, a formula like
{x}E-12 does not work when it comes to attempting the question. It needs to be written as {x}*1E-12, which seems far more sensible to me anyway.
By the way, a formula like
{x}E-12 COULD work when it comes to attempting the question.
The {x}
will be converted to the numerical to which the exponent will be applied as E number style is allowed in PHP
2.4E3 = 2400
I use this here to test the conversion process as any operation (ie.
*1E-12 ) could interfere with what we need to test.
P.S. even generating a number as large as in 1E45 range give a valid result.
992964467961790114603370108662767761814978560.0E3 = 9.9e47
PHP math functions seem quite robust..
Integrated (blind mode). Thanks!
Tested. Passed. Thanks 
fixes have been rolled merrily up the stream! Thanks everybody!
Would be nice to see some additional unit tests to cover this case.