Bug 239897 – [GCC] REGRESSION(r293605): error: cannot convert ‘<brace-enclosed initializer list>’ to ‘unsigned char:3’ in initialization

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Diego Pino]

The Debian 10 build bot (GCC 8.3) is failing with the following error:

```
./../Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h:113:43: error: cannot convert ‘<brace-enclosed initializer list>’ to ‘unsigned char:3’ in initialization
```

https://build.webkit.org/#/builders/46/builds/13117

It seems setting the bit size of a typed enum is not supported by GCC8.3.

[Diego Pino]

Created attachment 458583 [details]
Patch

[Darin Adler]

Comment on attachment 458583 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=458583&action=review

> Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h:111
> -    Kind m_kind : bitsOfKind { Kind::None };
> +    Kind m_kind { Kind::None };

I believe this will make the object larger. Can we keep the optimization under other compilers?

[Yusuke Suzuki]

Comment on attachment 458583 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=458583&action=review

>> Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h:111
>> +    Kind m_kind { Kind::None };
> 
> I believe this will make the object larger. Can we keep the optimization under other compilers?

Yes, this needs to be small. So, we need different solution.

[Darin Adler]

Comment on attachment 458583 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=458583&action=review

>>> Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h:111
>>> +    Kind m_kind { Kind::None };
>> 
>> I believe this will make the object larger. Can we keep the optimization under other compilers?
> 
> Yes, this needs to be small. So, we need different solution.

Should be able to work around this by using the type unsigned and casting to and from Kind when accessing the data member. It would be a shame to do that forever though. We want to work around the problem somehow in a way that can easily be removed once gcc is sufficiently improved.

[Diego Pino]

Created attachment 458842 [details]
Patch

[Yusuke Suzuki]

Comment on attachment 458842 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=458842&action=review

r=me with comment.

> Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h:50
> +    MethodOfGettingAValueProfile() { m_kind = Kind::None; }

Please use initialization list, and annotate it with constexpr.

constexpr MethodOfGettingAValueProfile()
    : m_kind(Kind::None)
{
}

[Diego Pino]

Created attachment 458852 [details]
[fast-cq] Patch

[Diego Pino]

(In reply to Yusuke Suzuki from comment #6)
> Comment on attachment 458842 [details]
> Patch

> Please use initialization list, and annotate it with constexpr.
> 
> constexpr MethodOfGettingAValueProfile()
>     : m_kind(Kind::None)
> {
> }

It seems the constructor cannot be `constexpr`.

```
/app/webkit/Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h: In constructor ‘constexpr JSC::MethodOfGettingAValueProfile::MethodOfGettingAValueProfile()’:
/app/webkit/Source/JavaScriptCore/bytecode/MethodOfGettingAValueProfile.h:51:28: error: call to non-‘constexpr’ function ‘JSC::CodeOrigin::CodeOrigin()’
   51 |         : m_kind(Kind::None)
```

[Diego Pino]

Created attachment 458855 [details]
Patch

[EWS]

Committed r293828 (250301@main): <https://commits.webkit.org/250301@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 458855 [details].

[Radar WebKit Bug Importer]

<rdar://problem/92789338>