Fix path completion when UTF8 char is occurring before match characters.

[dhoegh]

Small fix to #8838 so completion works again on: "β $dir_space\\space", @tkelman, @ivarne

[ivarne]

Looks good to me.

[stevengj]

Don't you have to handle the m == nothing case?

See also my comment. It's not clear to me why the original code is incorrect, since m.offset is the offset of an ASCII character (a single code unit).

[dhoegh]

I do not handle nothing because the regex should be garented a match due to the ´if´ at line 189.
About the ´nextind´

julia>s="α C:/"
julia>start = match(r" ",reverse(s)).offset
julia>s[length(s)-start]
'α'

So I need to step past the UTF-xx char. It could be changed to:

nextind(partial, length(partial) - m.offset) + 1

because I know the match char is ASCII.

[stevengj]

Doesn't s[endof(s) - start + 1] work fine? Hmm, no, that's not right either because endof gives the last character index, not the last byte index. Seems like it should be s[nextind(s,endof(s)) - start]

The real problem here seems to be the use of length instead of endof, which is wrong because offset is a byte index not a character index. e.g. your patch gives the wrong result 'β' for s="αβ C:/αβ"; start = match(r" ",reverse(s)).offset; s[nextind(s, nextind(s, length(s) - start))]

[stevengj]

It seems like searching strings in reverse is common enough and so error-prone that we should export a reverseind(s,i) = nextind(s,endof(s)) - i function from Base, and document it along with reverse. Although the implementation would be more complicated if i is not the index of an ASCII character.

[dhoegh]

@stevengj thank you for the pointer to endof, it was wrong to use length. But it still seems like it should be start = nextind(s, endof(s) - match(r" ",reverse(s)).offset) + 1 to get from C.

julia>s="αβ C:/αβ"
julia>start = nextind(s, endof(s) - match(r" ",reverse(s)).offset) + 1
julia>s[start:end]
"C:/αβ"

I have updated my comment

[stevengj]

nextind(s,endof(s)) - offset gives the index of the matched (ASCII) character in s. If you want the index of the character after that, you can use nextind(s,endof(s)) - offset + 1 because the matched character is ASCII.

Your code does not work if the character before the matched character is ASCII:

julia> s="αβd C:/αβ";
julia> start = nextind(s, endof(s) - match(r" ",reverse(s)).offset) + 1;
julia> s[start:end]
" C:/αβ"

(Notice that the matched character is included in this case, whereas it was not included for s="αβ C:/αβ".)

[dhoegh]

@stevengj I have found the solution, it converts to charecter index and the back:

julia> s="αβ C:/αβ";
julia> start = chr2ind(s, length(s) - ind2chr(reverse(s), match(r" ",reverse(s)).offset) + 2);
julia> s[start:end]
"C:/αβ"

I have updated.
I fogot to reverse the string

[stevengj]

My solution is faster: ind2chr and chr2ind are O(n) operations. Why don't you want to use nextind(s,endof(s)) - offset + 1?

[dhoegh]

@stevengj I have updated it with you proposal, and added a UTF-xx string to the path. Sorry I did not realized it worked you solution worked, I was pretty busted when I looked at it yesterday.

[stevengj]

LGTM, thanks!

[dhoegh]

It should be backported @JuliaBackports.

[ivarne]

Backported in 967b37a