[PATCH v8 04/11] jscript: Use string hint when retrieving the default value primitive in IE10+.

[Gabriel Ivăncescu]

On 26/11/2021 17:37, Jacek Caban wrote:
> On 11/26/21 12:52 PM, Gabriel Ivăncescu wrote:
>> On 25/11/2021 22:11, Jacek Caban wrote:
>>> Hi Gabriel,
>>>
>>> On 11/25/21 3:00 PM, Gabriel Ivăncescu wrote:
>>>> diff --git a/dlls/mshtml/tests/documentmode.js 
>>>> b/dlls/mshtml/tests/documentmode.js
>>>> index 08fcde4..c2fd8bd 100644
>>>> --- a/dlls/mshtml/tests/documentmode.js
>>>> +++ b/dlls/mshtml/tests/documentmode.js
>>>> @@ -1136,7 +1136,6 @@ sync_test("elem_attr", function() {
>>>>       ok(r === (v < 8 ? arr : (v < 9 ? "arrval" : null)), "testattr 
>>>> with custom valueOf = " + r);
>>>>       elem.setAttribute("testattr", arr);
>>>>       r = elem.getAttribute("testattr");
>>>> -    todo_wine_if(v >= 10).
>>>>       ok(r === (v < 8 ? arr : (v < 10 ? "arrval" : "42")), "testattr 
>>>> after setAttribute with custom valueOf = " + r);
>>>>       ok(elem.testattr === arr, "elem.testattr after setAttribute 
>>>> with custom valueOf = " + elem.testattr);
>>>>       r = elem.removeAttribute("testattr");
>>>
>>>
>>> This test seems too little to back such change. For example, another 
>>> likely explanation for the observed behaviour is that setAttribute 
>>> does explicit conversion to string with proper hints.
>>>
>>>
>>
>> How should I do that? Looking up toString like I had originally, or 
>> something else? Or do I need an internal interface? (in which case, 
>> I'll have to integrate it somehow to the proxy patches...) 
> 
> 
> As I suggested earlier, look at IVariantChangeType. change_type() should 
> already do what you need. The only problem is that you don't have a 
> caller context in setAttribute(). It's, however, available to hooks, so 
> maybe we could do the conversion in a hook.
> 
> 
> Thanks,
> 
> Jacek
> 

Ah, I see now, thanks for the pointer, that makes sense.

For the rest of the series, I think I'll postpone the patches that use 
the generic hooks though, I've discovered some interesting things with 
quirks/compat constructors and prototypes (which aren't in jscript but 
aim to "emulate" it somehow), and I believe now that those behaviors are 
some side-effect of that.

I'm not sure yet if it works that way, but it might make the 
implementation cleaner, so I'll refactor it after the proxy patches. 
I'll try to see if I can just get the other parts in (like the 
stringifying for get/setAttribute) without breaking existing tests, to 
avoid piling too many...