Fix: String#gsub replacing ascii chars to work in non-ascii string

[straight-shoota]

Fixes #5348

[straight-shoota]

This is a fix for the immediate issue.
It would probably be better if gsub_ascii_char worked with non-ascii strings, it is still just replacing an ascii char for another.

UPDATED: I've implemented the better solution.

[straight-shoota]

This would do as a proper solution in gsub_ascii_char, but I'm not fully confident about that pointer copying:

if my_char == char
  buffer[i] = replacement.ord.to_u8
else
  (buffer + i).copy_from(to_unsafe + i, my_char.bytesize)
end

[larubujo]

solution is to use each_byte_with_index, not go by char when char is ascii. faster and better. no need for string to be full ascii.

[straight-shoota]

There is no each_byte_with_index, but to_slice.each_with_index will do.

[straight-shoota]

Can I get a review? It should be a good solution now, not just a workaround as initially announced.

[RX14]

Please don't use ===, use char.ord == byte.

[RX14]

Also i'd much prefer

if char.ord == byte
  buffer[i] = replacement.ord.to_u8
else
  buffer[i] = byte
end

to avoid the ugly indent.

[Sija]

@RX14 I vote to change formatter rule for this indent to uniformly used 2 spaces.

[RX14]

That's another issue, and I disagree with you.

[Sija]

Out of the curiosity, how's that you don't want to change the rule, while in the same time you suggest against using it? Is there sth I don't get?

[RX14]

Because changing the rule would mean it looks like

buffer[i] = if char.ord == byte
  replacement.ord.to_u8
else
  byte
end

which i think is even more ugly than indenting it all.

My preference goes:

1. what I suggested (assign in the if)
2. what's implemented in this PR (indent the if)
3. what you suggested (don't indent the if)

[Sija]

Got it! for my taste it'd be 3,1,2 - in that order.

[straight-shoota]

2. is what it currently looks like. Until now this PR doesn't change anything about that.

[RX14]

well it won't get changed if it's not in this PR

[straight-shoota]

@RX14 now it is =)

[chastell]

make spec passes for me on this PR, so the TCPSocket failures on Travis are probably unrelated; could someone with access restart the Travis jobs?

[RX14]

This PR needs to be rebased onto master to pick up the latest travis.yml changes to pass CI.

[straight-shoota]

Rebased.

[RX14]

@bcardiff this should also probably be cherry-picked into 0.24.2 but i've milestoned it Next until that happens.