Bug 199831 – Add DataMutex and MainThreadData wrappers

• Component: Web Template Framework
• Status: RESOLVED FIXED

[Alicia Boya García]

DataMutex<T> stores an instance of T in a private member along with a
mutex. In order to use its fields, users need to instantiate a
DataMutex<T>::LockedWrapper instance in the stack. This class uses
RAII to lock and unlock the mutex in construction and destruction
respectively, and by using the arrow operator lets the user access T's
members.

This way, DataMutex<T> prevents most instances of accidental access to
data fields that should only be read and modified in an atomic matter.

Still, both the Lock and the LockHolder are exposed once the user has
taken the lock so that special cases such as waiting for a condition
variable or performing an early unlock are doable.

MainThreadData<T> is another wrapper class, in this case for data
fields that should only be accessed from the main thread. In this
case, it works similar to a smart pointer, except that (1) there is no
actual memory indirection, T is stored directly inside
MainThreadData<T> and (2) attempts to use the -> or * operator have an
isMainThread() assertion.

Together, these two wrapper classes make it easier to write
multi-threaded code in a safer, more self-documented way by letting
the author group data into structures that have certain access safety
requirements.

These structures were originally part of the new GStreamer
WebKitMediaSrc rework patch: https://bugs.webkit.org/show_bug.cgi?id=199719

[Alicia Boya García]

Created attachment 374214 [details]
Patch

[Alex Christensen]

Comment on attachment 374214 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=374214&action=review

> Source/WTF/wtf/MainThreadData.h:29
> +class MainThreadData {

This feels like it will want an operator T&()

[Alicia Boya García]

Comment on attachment 374214 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=374214&action=review

>> Source/WTF/wtf/MainThreadData.h:29
>> +class MainThreadData {
> 
> This feels like it will want an operator T&()

What would it do?

[Alex Christensen]

Comment on attachment 374214 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=374214&action=review

>>> Source/WTF/wtf/MainThreadData.h:29
>>> +class MainThreadData {
>> 
>> This feels like it will want an operator T&()
> 
> What would it do?

return a reference to m_data

[Alicia Boya García]

Comment on attachment 374214 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=374214&action=review

>>>> Source/WTF/wtf/MainThreadData.h:29
>>>> +class MainThreadData {
>>> 
>>> This feels like it will want an operator T&()
>> 
>> What would it do?
> 
> return a reference to m_data

That would make it easier to store it and pass it away, allowing for unprotected access. I want to make that uncomfortable on purpose :)

If someone really really wants, they can use &*, in a similar way to how you can get an unprotected pointer from a shared_ptr<>.

[Alex Christensen]

ok

[WebKit Commit Bot]

Comment on attachment 374214 [details]
Patch

Clearing flags on attachment: 374214

Committed r247723: <https://trac.webkit.org/changeset/247723>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/53446150>