Version: (using KDE Devel) Installed from: Compiled sources OS: NetBSD Bug was discovered while debuging kprinter crash on NetBSD - it always crashed at startup in getservbyname_r. Recent NetBSD releases don't provide reentrant (*_r) function declarations in headers - getservbyname, getservbyport etc. are already reentrant. There is nothing wrong with this and there is no any reason to bug NetBSD developers - *_r function declarations were removed quite recently from public headers in NetBSD. However, *_r function symbols exist in libc for backward binary compatibility. So, in while compiling KDE HAVE_GETSERVBYNAME_R will be set to 1 on NetBSD because symbol exists in libc, but HAVE_DECL_GETSERVBYNAME_R will be 0, because there is no declaration in public headers. KDE provides own getservbyname_r, getservbyport_r, getprotobyname_r and getprotobynumber_r in such case (kdecore/network/kresolver_p.h). Problem is that these declarations are wrong for NetBSD. NetBSD *_r API is similar (if not the same, I'm not familiar enough with internals of these platforms to say that) to the one used in case of OpenBSD - see for example http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libc/net/getservbyname_r.c?rev=1.3. IMHO the logic used in kdecore/network/kresolver_p.h is just plain wrong. It can't be assumed that if symbol exists then declaration provided there should work. Is there any more platforms which use this assumption at the moment? For NetBSD this logic should work - if *_r declarations exist, these should be used, if these don't exist, then it should be assumed that getservbyname etc are already reentrant. I volunteer to work on solution and provide patches (both 3.5 and trunk are affected of course), but I'd like to have discussion at first what direction to go.
I thought we had fixed that recently in Subversion.
I don't see any progress on this. Does it mean that there will be no solution in KDE3?
SVN commit 648568 by thiago: Add the code for KSocketFactory. For now, it doesn't support any proxies. I'll add the code for that later. Let's see if it breaks compilation anywhere. Also, use this opportunity to fix an old-time NetBSD problem: it provides reentrant a getservbyname() function. So assume all NetBSDs running KDE 4 have it reentrant. BUG:127253 M +1 -0 CMakeLists.txt AM network/ksocketfactory.cpp [License: LGPL (v2+)] M +6 -0 network/ksocketfactory.h --- trunk/KDE/kdelibs/kdecore/CMakeLists.txt #648567:648568 @@ -186,6 +186,7 @@ network/kstreamsocket.cpp network/ksocketbuffer.cpp network/klocalsocket.cpp + network/ksocketfactory.cpp network/netsupp.cpp localization/kcatalog.cpp localization/kcharsets.cpp ** trunk/KDE/kdelibs/kdecore/network/ksocketfactory.cpp #property svn:eol-style + native --- trunk/KDE/kdelibs/kdecore/network/ksocketfactory.h #648567:648568 @@ -23,9 +23,11 @@ #include <kdecore_export.h> #include <QtCore/QString> +#include <QtNetwork/QNetworkProxy> class QTcpSocket; class QTcpServer; +class QUdpSocket; class QHostAddress; /** @@ -164,6 +166,10 @@ static QTcpServer *listenAt(const QString &service, int defaultPort = -1, QObject *parent = 0); static QUdpSocket *datagramSocket(const QString &node, const QString &service, QObject *parent = 0); + + static QNetworkProxy proxyForConnection(const QString &node, const QString &service); + static QNetworkProxy proxyForListening(const QString &service); + static QNetworkProxy proxyForDatagram(const QString &node, const QString &service); }; #endif