replace UnicodeException with UnicodeError

[MartinNowak]

• first step to make invalid UTF encodings in strings
  a programming error (ensuring valid encoding has to be
  done when converting raw input data to string)
• attributes for rt.util.utf
• added staticError helper to throw @nogc Errors

[MartinNowak]

The outcome of Issue 14519 – [Enh] foreach on strings should return replacementDchar rather than throwing is that the programmer has to ensure that strings contains valid unicode data (e.g. by validating raw data as done by readText) and any algorithm working on strings should assert that but not repeat the validation.

[CyberShadow]

readText calls validate, which used to throw UnicodeException, so won't this break validating text in readText?

[MartinNowak]

You're right, it's a different validate, but this one shouldn't be nothrow and we still need UTFExcwption for validation.
BTW the signature of validate should become inout(char)[] validate(input(ubyte)[] str);.

[CyberShadow]

I'm still unsure about the whole ubyte-for-ASCII idea. I think it's an interesting direction iff we improve Phobos and the language to better support it as well (e.g. allowing comparison of ubyte[] and string literals, making s.startsWith("foo") work with ubyte[] s, etc.)

[schuetzm]

What do you mean by "ubyte-for-ASCII"? ASCII is a subset of UTF8, char is fine for it.

[CyberShadow]

Sorry, I meant non-UTF 8-bit encodings.

[MartinNowak]

Well s.startsWith("foo".representation) would work.
You can't generally mix string with ubyte[] because one needs to use autodecoding, while the other can't.

[DmitryOlshansky]

added staticError helper to throw @nogc Errors

Useful in and out of itself. Break it out?

[MartinNowak]

Useful in and out of itself. Break it out?

Done, #1325.

[d-random-contributor]

void validate(S)(in S s) pure @safe
{
  if (!s.invalidChars.empty)
    throw new UnicodeException();
}

auto invalidChars(S)(in S s) pure nothrow @safe @nogc;

Do we have nothrow @nogc validation?

[DmitryOlshansky]

Please rebase, #1325 was pulled.

[DmitryOlshansky]

Any updates on this? @MartinNowak

[MartinNowak]

It's really to wrong approach to start this transition by replacing some exceptions in druntime.
I'll try to bring this topic up for discussion again, hopefully we can reach a consensus.