OK, this exercise has some complicated issues...

1. The canonical data states "implementors should use proper fixed-point or currency data types of the corresponding language and not float." Our implementation currently uses float. We may opt to use int and cents as the canonical test data contains them already or continue using float.
2. assertEquals() is not appropriate for float values. There is assertEqualsWithDelta() for floating point comparison. This might already make the current example code pass the tests.
3. The missing tests seem to be those that "optimize" the price for cases with 5 different books but a group of 5 is not the optimum. I do not know the history of the exercise, but these seem to be added to catch edge cases of optimizing. I think we should have them.

Any additional findings / thoughts?

My 5cents.

1. I would go for int as that is the right way to handle amount in PHP, if not using a money library.

2. The assertEqualsWithDelta() would not be needed then.

3. I think we should include all test cases, en ever exercise, unless there are some functionality in PHP making it super hard or impossible.

I have forgotten have to look up how many people have already solved the exercise, and how many it would break for.

There is 61 solutions, only ~40% of 153 people who tried this. So this should be improved.

@homersimpsons Your opinion about 1. / 2.? Proper float comparison or using int instead?

@homersimpsons Your opinion about 1. / 2.? Proper float comparison or using int instead?

I know I always prefer int for monetary values. I think it is a good way to show students that this could be used and avoids most rounding issues. It may helps them in the future too when they will handle monetary values.

I would prefer to avoid float here. Another alternative could be to use BCMath.

So my vote is (from most preferred to least preferred) int > bcmath >> float.

There is 61 solutions, only ~40% of 153 people who tried this. So this should be improved.

61 solutions looks fine for me to change this completely.

OK, thanks everybody. We agreed on int and as we break existing solutions with that we will have all test cases with exactly the data from canonical data. @tomasnorre please continue with the PR.

I'll take care tomorrow or beginning of next week.