Bug 213449 – [WTF] Ensure ASCIILiteral is ASCII characters at compile time

• Component: New Bugs
• Status: RESOLVED FIXED

[Yusuke Suzuki]

[WTF] Ensure ASCIILiteral is ASCII characters at compile time

[Yusuke Suzuki]

Created attachment 402438 [details]
Patch

[Ross Kirsling]

Comment on attachment 402438 [details]
Patch

r=me. ラーメン食べたい。

[Mark Lam]

Comment on attachment 402438 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=402438&action=review

> Source/WTF/wtf/text/ASCIILiteral.h:64
> +        ASSERT_UNDER_CONSTEXPR_CONTEXT(isASCII(characters[i]));

Since this is constexpr, is there a reason why we only want this for ASSERT_ENABLED?  Why not have RELEASE_ASSERT_UNDER_CONSTEXPR_CONTEXT this?

[Yusuke Suzuki]

Comment on attachment 402438 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=402438&action=review

>> Source/WTF/wtf/text/ASCIILiteral.h:64
>> +        ASSERT_UNDER_CONSTEXPR_CONTEXT(isASCII(characters[i]));
> 
> Since this is constexpr, is there a reason why we only want this for ASSERT_ENABLED?  Why not have RELEASE_ASSERT_UNDER_CONSTEXPR_CONTEXT this?

Discussed with Mark on Slack, this function can be called in non-constexpr context, like,

auto string = "Hey"_s; // instead of constexpr string = "Hey"_s; in non global code.

So, this is safer for accidental runtime checks.

[EWS]

Committed r263341: <https://trac.webkit.org/changeset/263341>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 402438 [details].

[Radar WebKit Bug Importer]

<rdar://problem/64585669>

[Adrian Perez]

Comment on attachment 402438 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=402438&action=review

> Source/WTF/ChangeLog:12
> +            auto globalVariable = "ã©ã¼ã¡ã³ (Ramen in Japanese)"_s;

As a ramen enthusiast, I like this example in the commit log ^_^ 🍜️♥️

[Darin Adler]

Very very happy to see this change go in.

[Darin Adler]

Is there some way to make this even more strongly guaranteed to be done at compile time? I assume the answer is no, but obviously we want to compile errors with a poorly formed literal, not a runtime check if it happens to be used in a non-constexpr context.

[Yusuke Suzuki]

(In reply to Darin Adler from comment #9)
> Is there some way to make this even more strongly guaranteed to be done at
> compile time? I assume the answer is no, but obviously we want to compile
> errors with a poorly formed literal, not a runtime check if it happens to be
> used in a non-constexpr context.

This is the purpose of consteval in C++20!
https://en.cppreference.com/w/cpp/language/consteval

I can't wait to use C++20 :D