Single backslash problem at end of term in sigil_S/2

[Eiji7]

Environment

$ asdf current
elixir         1.8.1-otp-21 (set by …)
erlang         21.3.6   (set by …)

Funtoo Linux current x86-64bit on intel64-ivybridge

Current behavior

Looks like that backslash (\) character is not properly handled at end of term in sigil_S/2. I have found it while trying sigil_S in iex shell, but this could affect other sigils as well.

For sure, here is current documentation of sigil_S/2:

iex(1)> h sigil_S

                       defmacro sigil_S(term, modifiers)                        

Handles the sigil ~S for strings.

It simply returns a string without escaping characters and without
interpolations.

## Examples

    iex> ~S(foo)
    "foo"
    
    iex> ~S(f#{o}o)
    "f\#{o}o"

Now let's see some examples:

# 1st - only " character:
~S(") # returns: "\"" as expected

# 2nd - only \ character:
~S(\)
) # returns: ")\n" ???

# 3rd - \" characters:
~S(\") # returns: "\\\"" as expected

# 4rd - "\ characters:
~S("\)
) # returns: "\")\n" ???

# 5th - \\ characters:
~S(\\) # returns: "\\\\" as expected

I believe that current behavior is not expected and if so then we should add to documentation that \ character at end of term needs to be escaped.

Expected behavior

Examples: 1st, 3rd and 5th should not change as they are already working as expected.

# 2nd - only \ character:
~S(\) # should return: "\\"

# 4rd - "\ characters:
~S("\) # should return: "\"\\"

[josevalim]

The behaviour is correct. Once ypu use \), you are escaping the terminator, therefore you are not closing the sigil. It is exactly what happens with strings to. If you write "\", then the string is incomplete and you still have to close it.

[LostKobrakai]

@josevalim I just saw some questions around that on slack. It seems ~S cannot be used if the string ends in a single backslash, because it always escapes the delimiter – unless the character before the backslash is also a backslash, in which case the resulting string will consist of two backslashes.

So the first backslash escaped the second backslash from escaping the string delimiter, but it still becomes part of the final string. To me that sounds like ~S"\\" is not working as expected. Imo it would be consistent if it either is equal to "\\" or it is equal to (the incomplete) "\\\". It's equal to neither however.

To me the second option sounds like the most sane, because a trailing backslash does always escape the delimiter, no matter what character comes before.

[josevalim]

You are correct that the backslash should not be escaped in such cases.