Re: mshtml #2: Added HTMLWindow's IDispatch methods implementation.

[Robert Shearman]

Jacek Caban wrote:
> static HRESULT WINAPI HTMLWindow2_item(IHTMLWindow2 *iface, VARIANT *pvarIndex, VARIANT *pvarResult)
> diff --git a/dlls/mshtml/main.c b/dlls/mshtml/main.c
> index 3f8131c..41f32eb 100644
> --- a/dlls/mshtml/main.c
> +++ b/dlls/mshtml/main.c
> @@ -51,10 +51,52 @@ DWORD mshtml_tls = 0;
>
> static HINSTANCE shdoclc = NULL;
>
> +static ITypeLib *typelib;
> +static ITypeInfo *typeinfos[LAST_tid];
> +
> +static const REFIID tid_ids[] = {
> +    &IID_IHTMLWindow2
> +};
> +
> +HRESULT get_typeinfo(enum tid_t tid, ITypeInfo **typeinfo)
> +{
> +    HRESULT hres;
> +
> +    if(!typelib) {
> +        hres = LoadRegTypeLib(&LIBID_MSHTML, 4, 0, LOCALE_SYSTEM_DEFAULT, &typelib);

This isn't thread-safe.
> +        if(FAILED(hres)) {
> +            ERR("LoadRegTypeLib failed: %08x\n", hres);
> +            return hres;
> +        }
> +    }
> +
> +    if(!typeinfos[tid]) {
> +        hres = ITypeLib_GetTypeInfoOfGuid(typelib, tid_ids[tid], typeinfos+tid);
> +        if(FAILED(hres)) {
> +            ERR("GetTypeInfoOfGuid failed: %08x\n", hres);
> +            return hres;
> +        }
> +    }

Why not just use the IID instead of an enumeration here? I know you're 
trying to cache the typeinfo here, but the current implementation of 
typelibs makes ITypeLib_GetTypeInfoOfGuid very cheap.
> +
> +    *typeinfo = typeinfos[tid];
> +    return S_OK;
> +}
> +
>  static void thread_detach(void)
>  {
> -    thread_data_t *thread_data = get_thread_data(FALSE);
> +    thread_data_t *thread_data;
> +
> +    if(typelib) {
> +        unsigned i;
> +
> +        for(i=0; i < sizeof(typeinfos)/sizeof(*typeinfos); i++)
> +            if(typeinfos[i])
> +                ITypeInfo_Release(typeinfos[i]);
> +
> +        ITypeLib_Release(typelib);
> +    }

Why are you doing this from DLL_THREAD_DETACH and not DLL_PROCESS_DETACH?

-- 
Rob Shearman

mshtml #2: Added HTMLWindow's IDispatch methods implementation.

[Jacek Caban]

Hi Robert,

Thanks for your review.

Robert Shearman wrote:
> Jacek Caban wrote:
> > static HRESULT WINAPI HTMLWindow2_item(IHTMLWindow2 *iface, VARIANT 
> > *pvarIndex, VARIANT *pvarResult)
> > diff --git a/dlls/mshtml/main.c b/dlls/mshtml/main.c
> > index 3f8131c..41f32eb 100644
> > --- a/dlls/mshtml/main.c
> > +++ b/dlls/mshtml/main.c
> > @@ -51,10 +51,52 @@ DWORD mshtml_tls = 0;
> >
> > static HINSTANCE shdoclc = NULL;
> >
> > +static ITypeLib *typelib;
> > +static ITypeInfo *typeinfos[LAST_tid];
> > +
> > +static const REFIID tid_ids[] = {
> > +    &IID_IHTMLWindow2
> > +};
> > +
> > +HRESULT get_typeinfo(enum tid_t tid, ITypeInfo **typeinfo)
> > +{
> > +    HRESULT hres;
> > +
> > +    if(!typelib) {
> > +        hres = LoadRegTypeLib(&LIBID_MSHTML, 4, 0, 
> > LOCALE_SYSTEM_DEFAULT, &typelib);
>
> This isn't thread-safe.

I don't consider it really important here ATM. I haven't seen any app 
that would benefit from thread safe MSHTML. Anyway I will fix it.
> > +        if(FAILED(hres)) {
> > +            ERR("LoadRegTypeLib failed: %08x\n", hres);
> > +            return hres;
> > +        }
> > +    }
> > +
> > +    if(!typeinfos[tid]) {
> > +        hres = ITypeLib_GetTypeInfoOfGuid(typelib, tid_ids[tid], 
> > typeinfos+tid);
> > +        if(FAILED(hres)) {
> > +            ERR("GetTypeInfoOfGuid failed: %08x\n", hres);
> > +            return hres;
> > +        }
> > +    }
>
> Why not just use the IID instead of an enumeration here? I know you're 
> trying to cache the typeinfo here, but the current implementation of 
> typelibs makes ITypeLib_GetTypeInfoOfGuid very cheap.

It's surely not cheaper than a simple table element check. It's a part 
of code that is extremely profiled in Gecko (that is its counterpart in 
Gecko). We don't have to take that much care of it, but I think such 
simple caching is wort the efforts.
> > +
> > +    *typeinfo = typeinfos[tid];
> > +    return S_OK;
> > +}
> > +
> >  static void thread_detach(void)
> >  {
> > -    thread_data_t *thread_data = get_thread_data(FALSE);
> > +    thread_data_t *thread_data;
> > +
> > +    if(typelib) {
> > +        unsigned i;
> > +
> > +        for(i=0; i < sizeof(typeinfos)/sizeof(*typeinfos); i++)
> > +            if(typeinfos[i])
> > +                ITypeInfo_Release(typeinfos[i]);
> > +
> > +        ITypeLib_Release(typelib);
> > +    }
>
> Why are you doing this from DLL_THREAD_DETACH and not DLL_PROCESS_DETACH?

Ops, that's my fatal mistake.


Thanks,
    Jacek