string interpolation and escaping bug

[StefanKarpinski]

julia> println("\\ $('x')")
 x

julia> println("\\ x")
\ x

[JeffBezanson]

Might be my fault. In the parser I unescape \ and " in strings that will be passed to macros. I don't have any idea why I'm doing that.

[StefanKarpinski]

Other symptoms:

julia> "\\\\"
"\\\\"

julia> S"\\\\"
"\\"


julia> "\\\""
"\\\""

julia> S"\\\""
"\""

[StefanKarpinski]

I think that's what's going on: you should turn \" into " but not turn \\ into \. Leaving \" as-is is harmless w.r.t. standard escaping rules, however, since \" just decodes as ". This is sort of mind-numbing stuff. Not quite sure what the right course of action is.

[StefanKarpinski]

Btw, I'm not sure about this at all. I'm still mulling over what the right thing to do here is.

[JeffBezanson]

I changed the parser behavior and it seems to work, and all tests still pass. I can't imagine why the behavior I had before was the right thing. Unless we can figure that out, we seem to be ok now.

[JeffBezanson]

Ah, there is the case of L"\"". I guess that should be a string of one quote character?
Then there is kind of a strange situation where you can't make a string of a single \ using L syntax. L"\" escapes the quote and is incomplete input, and L"\\" is two backslashes.

[StefanKarpinski]

I'm kind of thinking that the behavior of d443299 is the best. True, you can't use L" to make just " or \, but I think that may be unavoidable. The simplest behavior to understand seems to be that L" always makes a string literal that's exactly what's between the opening and closing quotes — and \" is a way of indicating that an intervening " is not the closing one. I'm pretty sure you can't avoid at least having your escape character \ be inexpressible in L" strings — this is why escaping exists. The only question is whether we want the same to apply to " or not. The difference between \ and " is that \ will be the escape character no matter what quote syntax exists, whereas " is specific to this quoting form. If we implement L""" then " by itself won't be a special character anymore. However, I still think the policy of passing the literal contents through may be best.

[JeffBezanson]

OK finally settled :)