We do not build texteditor files in unit-tests so some tricks
were required to make ClangFormatIndenter available.
First simple unit-test proofs it builds and runs.
Change-Id: I81d5ea099bd27fd1c1ed8b5b7877299dcc62a67f
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@qt.io>
The group identifier is as good as the actual object, so use that
on the generator and the consumer side.
Change-Id: I6ccd54a2362631af0b50583f12563b8e41a5dd78
Reviewed-by: David Schulz <david.schulz@qt.io>
... by additionally keeping local (currently non-owning) pools per
"interesting" type.
Current situation:
- The global object pool does not scale well for looking up
objects, as iteration plus qobject_cast typically iterates
over all pooled objects.
- User code that can use typed results from the object
pool need to have access to the full type definition anyway,
i.e. depend on the plugin of the target class anyway.
The patch here solves the scaling problem is to have local
type-specific pools to which objects register in their
constructors and deregister in their destructors.
This patch here does *not* change the ownership model of the
pooled objects, however, it opens the possibility to change
the ownership model per type (e.g. by not putting things into
the global pool at all anymore and make the local pool 'owning')
and the intent is to handle that in later patchs.
Even without the follow-up patches this here is a performance
improvement for the cases that access the local pools instead
the global one, i.e. "practically all".
Change-Id: Ib11a42df2c4ecf5e1155534730083a520dd1995b
Reviewed-by: Eike Ziller <eike.ziller@qt.io>
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
