Regex.split with trailing \K works incorrectly

[error256]

Elixir 1.18.3, Erlang/OTP 27.3.3

IO.inspect Regex.split ~r/b\K/, "ababab"
IO.inspect :re.split "ababab", "b\\K"

returns

["ab", "ab", "", "ab", "", ""]
["ab", "ab", "ab", ""]

I think Regex.split should return the same result as :re.split. Non-trailing \Ks and lookbehinds seem to work correctly, this is the only case.

[metemaad]

You can directly use :re.split till the fix is ready

…

On Sat, May 3, 2025 at 5:41 PM error256 ***@***.***> wrote: *error256* created an issue (elixir-lang/elixir#14467) <#14467> Elixir 1.18.3, Erlang/OTP 27.3.3 IO.inspect Regex.split ~r/b\K/, "ababab"IO.inspect :re.split "ababab", "b\\K" returns ["ab", "ab", "", "ab", "", ""] ["ab", "ab", "ab", ""] I think Regex.split should return the same result as :re.split. Non-trailing \Ks and lookbehinds seem to work correctly, this is the only case. — Reply to this email directly, view it on GitHub <#14467>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGJSGFKLY7JQHXCFIKTQCYD24UZZLAVCNFSM6AAAAAB4MEOPCWVHI2DSMVQWIX3LMV43ASLTON2WKOZTGAZTONRSGY2TONQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[josevalim]

Here is the relevant Erlang code that deals with this and we are missing, in particular the first branch of the case:

do_split(Subj, Offset, [[{MainI,MainL}|Sub]|T], NumSub, Limit, Group) ->
    NewOffset = MainI+MainL,
    KeptLen =  MainI - Offset,
    case {KeptLen,empty_sub(Sub),MainL} of
	{0,true,0} ->
	    do_split(Subj,NewOffset,T,NumSub,Limit,Group);
	_ ->
	    <<_:Offset/binary,Keep:KeptLen/binary,_/binary>> = Subj,
	    ESub = extend_subpatterns(Sub,NumSub),
	    Tail = do_split(Subj, NewOffset, T, NumSub, Limit - 1,Group),
	    case Group of
		false ->
		    [Keep | dig_subpatterns(Subj,lists:reverse(ESub),Tail)];
		true ->
		    [[Keep | dig_subpatterns(Subj,lists:reverse(ESub),[])]|
		     Tail]
	    end
    end.
empty_sub([]) ->
    true;
empty_sub([{_,0}|T]) ->
    empty_sub(T);
empty_sub(_) ->
    false.

[sabiwara]

@josevalim I came to the same conclusion and tried a fix, but it breaks some tests so it might be a breaking change?

iex> :re.split "abc", ""
["a", "b", "c", ""]

# main
iex> Regex.split ~r//, "abc"
["", "a", "b", "c", ""]

# the fix
iex> Regex.split ~r//, "abc"
["a", "b", "c", ""]