Decimal.to_integer strangeness

kip

ex_cldr Core Team

I’m hoping smarter eyes than mine can help me resolve this issue. I am attempting to convert a decimal to an integer. The Decimal implementation uses guard clauses to decide if such as conversion is possible (good!). But for some reason I’m seeing guard clauses fail even when the data looks like it should pass. Here’s the example:

# A decimal. Shouldn't be able to be converted to an integer
# but it looks like the guards for the third clause should match and I
# don't know why they aren't!
iex> d = Decimal.new("1234.50")
#Decimal<1234.50>

# Here's what the struct underneath looks like:
iex> inspect d, structs: false
"%{__struct__: Decimal, coef: 123450, exp: -2, sign: 1}"

# Lets check the remainder of the coef
iex> Kernel.rem(d.coef,10)
0

# Error raises on conversion suggesting no guards match but
# I think the third clause should match?
iex> Decimal.to_integer d                     
** (FunctionClauseError) no function clause matching in Decimal.to_integer/1    
    
    The following arguments were given to Decimal.to_integer/1:
    
        # 1
        #Decimal<1234.5>
    
    Attempted function clauses (showing 3 out of 3):
    
        def to_integer(%Decimal{sign: sign, coef: coef, exp: 0}) when is_integer(coef)
        def to_integer(%Decimal{sign: sign, coef: coef, exp: exp}) when is_integer(coef) and exp > 0
        def to_integer(%Decimal{sign: sign, coef: coef, exp: exp}) when is_integer(coef) and exp < 0 and rem(coef, 10) == 0
    
    (decimal 2.0.0) lib/decimal.ex:1322: Decimal.to_integer/1

But in my example is_integer(coef) == true and exp < 0 == true and rem(coef, 10) == 0 so why isn’t this clause being executed?

Probably an issue with me staring at a screen for too long so any guidance appreciated.

5 comments

#decimal #troubleshooting

2 1519 5

2022-01-26 07:12:51 UTC

Showing Posts 1 to 5

Show Best Posts
Show All (oldest first)
Show All (newest first)

Aetherus

I copied and pasted all your statements into my Livebook, and I got the same error

Post #1

Aetherus

I did my experiment by copy and paste:

defmodule MyDecimal do
  # `to_integer/1` is copied from the source code of `Decimal`,
  # and added some `IO.inspect`

  def to_integer(%Decimal{sign: sign, coef: coef, exp: 0} = d)
      when is_integer(coef) do
        IO.inspect(d, label: "Branch 1", structs: false)
        sign * coef
      end

  def to_integer(%Decimal{sign: sign, coef: coef, exp: exp} = d)
      when is_integer(coef) and exp > 0 do
        IO.inspect(d, label: "Branch 2", structs: false)
        to_integer(%Decimal{sign: sign, coef: coef * 10, exp: exp - 1})
      end

  def to_integer(%Decimal{sign: sign, coef: coef, exp: exp} = d)
      when is_integer(coef) and exp < 0 and Kernel.rem(coef, 10) == 0 do
        IO.inspect(d, label: "Branch 3", structs: false)
        to_integer(%Decimal{sign: sign, coef: Kernel.div(coef, 10), exp: exp + 1})
      end
end

MyDecimal.to_integer(Decimal.new("1234.50"))

and it prints

Branch 3: %{__struct__: Decimal, coef: 123450, exp: -2, sign: 1}

** (FunctionClauseError) no function clause matching in MyDecimal.to_integer/1

The following arguments were given to MyDecimal.to_integer/1:

    # 1
    #Decimal<1234.5>

    #cell:2: MyDecimal.to_integer/1

So, the “problem” occurs when trying to do to_integer(%Decimal{sign: 1, coef: 12345, exp: -1}).

Maybe Decimal does not want you to do anything ambiguous about the fraction part.

Post #2

kip

ex_cldr Core Team

Thanks very much for following this through. And spotting my silly error - which is to not notice that the third clause is recursive (as it needs to be). I know that the number can’t be converted to an integer, I just couldn’t see why it wouldn’t match on the third clause. And indeed it does match on the first pass. I missed that it was recursing and not matching on a later pass (in fact there is an accepted PR that handles this error case better then the published version).

Thanks again.

Post #3

wojtekmach

Hex Core Team

I think latest Decimal should have a proper error message. Or maybe that change wasn’t released yet. If still have a bad error message, that’s definitely a bug and please report it!

Post #4