Re: Challenges of adding octal and hexadecimal escape sequences in strings

[Jose E. Marchesi]

>> > On Fri, Oct 30, 2020 at 05:51:44PM +0100, Jose E. Marchesi wrote:
>> >> 
>> >> I don't think that is the desired behavior.  If we define Poke strings
>> >> as NULL-terminated, it is obvious they are not to contain any NULL
>> >> byte.  Now, we can achieve that by two ways:
>> >> 
>> >> a) By doing what we do now, i.e to ignore any part of the string after a
>> >>    NULL character, in case one is inserted, or
>> >> 
>> >> b) To emit an error whenever a 0 byte is attempted to be stored in a
>> >>    string value.
>> >> 
>> >> Supposing we wanted to switch to b), at the moment the only ways to
>> >> create a string value in Poke are:
>> >> 
>> >> b1) Using a string literal, and
>> >> b2) using a cast from uint<8> to string
>> >> 
>> >> The first case is easy to implement: just emit a compile-time error if a
>> >> string literal contains a null character.
>> >> 
>> >> For the second case we have two choices: either to raise an exception
>> >> when `8UB as string' is executed, or to keep the current behavior of
>> >> generating an empty string "", but the latter would be an exception to
>> >> the rule.
>> >> 
>> >> That's why I prefer a) to b): it is more orthogonal, and can be
>> >> explained by a single rule without needing any exceptions to the rule.
>> >> Also, it comes handy to shorten strings :D
>> >
>> > What about a third way:
>> >
>> > c) Keep Poke strings as NULL-terminated. Treat using of NULL character in 
>> > string
>> >    literals as compilation error.
>> >
>> > The user cannot access data beyond the first NULL character, and having 
>> > bytes
>> > beyond that character doesn't make any sense.
>> 
>> Isn't that a subset of b) above?
>
> It's a nice subset of the "b" :D
> Just "b1".
>
> I mean if Poke string is a NULL-terminated thing, and user tries to make a
> string using a "string literal" (a compile-time thing) we should emit an 
> error.
> The other parts like "cast from uint<8> to string" is not important.
> Because the user (probably) knows what he/she is doing (and it's a runtime 
> thing).
> But for a "string literal", IMHO the user definitely doesn't know what he/she 
> is
> doing! (because he/she cannot access the bytes beyond the first NULL 
> character).
>
> ```poke
> defvar s = "a\0b"; // this should emit an compiler error (not an exception).
>                    // this should be considered as meaningless as
>                    // "defvar i = + 1 2;"
> ```
>
> I think "error" is better than "ignoring" for string literals.

Sure, I would be ok with having the compiler to emit an error in these
cases, but that doens't require a change on how strings are handled.