From kde-core-devel  Mon Apr 29 22:16:29 2013
From: Thiago Macieira <thiago () kde ! org>
Date: Mon, 29 Apr 2013 22:16:29 +0000
To: kde-core-devel
Subject: Re: Initialization of QSharedDataPointer's for empty objects (i.e. Foo::Foo(void))
Message-Id: <1647832.4WUlSLJ3Nl () tjmaciei-mobl2>
X-MARC-Message: https://marc.info/?l=kde-core-devel&m=136727382520250
MIME-Version: 1
Content-Type: multipart/mixed; boundary="--nextPart2502548.c8ExvUV4jW"


--nextPart2502548.c8ExvUV4jW
Content-Transfer-Encoding: 7Bit
Content-Type: text/plain; charset="us-ascii"

On segunda-feira, 29 de abril de 2013 21.52.37, Milian Wolff wrote:
> I.e.: In the ctors that construct an "empty" object do not call "new
> Private" but instead (re-)use a static shared empty object created on the
> stack - see shared_empty and shared_null in qstring.{cpp,h}.

You probably did not mean what you said: you do not want something allocated 
on the stack. You want a global. Like QString :-)

To be *really* like QString, you want your private to be POD. That means you 
cannot derive from QSharedData, you cannot use QAtomicInt, QString or 
QByteArray. That may be taking it a bit too far.

If your private class is not POD, you should use a K/Q_GLOBAL_STATIC instead 
to hold your private. Examples:

corelib/mimetypes/qmimedatabase.cpp:Q_GLOBAL_STATIC(QMimeDatabasePrivate, 
staticQMimeDatabase)
corelib/tools/qlocale_win.cpp:Q_GLOBAL_STATIC(QSystemLocalePrivate, 
systemLocalePrivate)
gui/text/qfontdatabase.cpp:Q_GLOBAL_STATIC(QFontDatabasePrivate, privateDb)
network/access/qabstractnetworkcache.cpp:Q_GLOBAL_STATIC(QNetworkCacheMetaDataPrivate, 
metadata_shared_invalid)
plugins/bearer/qnetworksession_impl.cpp:Q_GLOBAL_STATIC(QNetworkSessionManagerPrivate, 
sessionManager);
sql/kernel/qsqlquery.cpp:Q_GLOBAL_STATIC_WITH_ARGS(QSqlQueryPrivate, 
nullQueryPrivate, (0))

> I do wonder though why QString has both, a shared_empty and a shared_null
> object. A single one should suffice for most objects, no?

	QString().isNull() == true
	QString("").isNull() == false

> Does anyone have any experience with this? Are there any pitfalls or
> downsides to the approach taken by QString? I also thought about not
> initializing the QSaredDataPointer at all but that would require a lot of
> additional code when its being used to ensure its only accessed when it is
> initialized. I think the pattern applied in QString is much nicer in that
> regard.

It is, but the pattern of having a null d pointer also exists elsewhere in Qt.

For example, in QUrl, it's implemented manually. But there are also a few 
examples of doing that with QSharedDataPointer.

Take a look at QNetworkProxy or QUrlQuery:
template<> void QSharedDataPointer<QNetworkProxyPrivate>::detach()
{
    if (d && d->ref.load() == 1)
        return;
    QNetworkProxyPrivate *x = (d ? new QNetworkProxyPrivate(*d)
                               : new QNetworkProxyPrivate);
    x->ref.ref();
    if (d && !d->ref.deref())
        delete d;
    d = x;
}

And then:
QNetworkProxy::ProxyType QNetworkProxy::type() const
{
    return d ? d->type : DefaultProxy;
}

-- 
Thiago Macieira - thiago (AT) macieira.info - thiago (AT) kde.org
   Software Architect - Intel Open Source Technology Center
      PGP/GPG: 0x6EF45358; fingerprint:
      E067 918B B660 DBD1 105C  966C 33F5 F005 6EF4 5358

--nextPart2502548.c8ExvUV4jW
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: This is a digitally signed message part.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.19 (GNU/Linux)

iD8DBQBRfvFLM/XwBW70U1gRAvHzAJ96B6gbDrnevYUsjr9hJmPCNeF7NACcDyMu
by1GVNA9ur0EDZYWVpE40W0=
=xtDn
-----END PGP SIGNATURE-----

--nextPart2502548.c8ExvUV4jW--