A friend of mine sent me a link containing an accent, http://öpnvkarte.de/
Clicking on it opened "http://pnvkarte.de/" in my browser.

I made some other check, and it appears that this issue happens only with .de domains and with any type of accented character.

ciao
Riccardo

and that would be great if another method were added openUrl(const QUrl&
url) here and in the urlobject.
QUrl guarantees to store proper url.

2010/3/9 Jan Niehusmann <jan@gondor.com>

Hi,

I got the following bug report through the debian bts. Looking at the
source code, it seems like the following function in desktoputil.cpp is
not correct:

bool DesktopUtil::openUrl(const QString& url)
{
        QByteArray ascii = url.toAscii();
        if (ascii == url)
                return doOpenUrl(QUrl::fromEncoded(ascii));
        else
                return doOpenUrl(QUrl(url, QUrl::TolerantMode));
}


QString.toAscii is passing on latin1 characters unchanged (assuming no
QTextCodec has been set). So (ascii == url) is true even though the url
contains the character 'ö', which is then discarded by QUrl::fromEncoded.

Just calling doOpenUrl(QUrl(url, QUrl::TolerantMode)) without checking
for pure-ascii urls first does work. But of course, this essentially
reverts 9790593a9b0c72b335a5c724cd0deca443d8fa1e, so directly entering
percent-encoded URLs doesn't work any more.

I guess it's necessary to explicitely check if the url only contains
legal 7-bit characters to decide how to interpret it.

Regards,
Jan

Hi,

I got the following bug report through the debian bts. Looking at the
source code, it seems like the following function in desktoputil.cpp is
not correct:

bool DesktopUtil::openUrl(const QString& url)
{
        QByteArray ascii = url.toAscii();
        if (ascii == url)
                return doOpenUrl(QUrl::fromEncoded(ascii));
        else
                return doOpenUrl(QUrl(url, QUrl::TolerantMode));
}


QString.toAscii is passing on latin1 characters unchanged (assuming no
QTextCodec has been set). So (ascii == url) is true even though the url
contains the character 'ö', which is then discarded by QUrl::fromEncoded.

Just calling doOpenUrl(QUrl(url, QUrl::TolerantMode)) without checking
for pure-ascii urls first does work. But of course, this essentially
reverts 9790593a9b0c72b335a5c724cd0deca443d8fa1e, so directly entering
percent-encoded URLs doesn't work any more.

I guess it's necessary to explicitely check if the url only contains
legal 7-bit characters to decide how to interpret it.

Regards,
Jan

Indeed unnecessary conversions between string and url should be
prevented, but when a message contains an url within the message text,
the conversion QString -> QUrl must be done somewhere.

To keep the changes small, I didn't touch the signature of openUrl, but
instead just changed the if condition. What do you think about the
following patch? It just checks if the url only contains printable 7-bit
ascii characters (range 0x21-0x7e) instead of checking if toAscii does
modify the contents of the string.

Jan

diff --git a/src/desktoputil.cpp b/src/desktoputil.cpp
index 4bc08a0..a9c7145 100644
--- a/src/desktoputil.cpp
+++ b/src/desktoputil.cpp
@@ -27,6 +27,7 @@
 #include <QProcess>
 #include <QSysInfo>
 #include <QSet>
+#include <QRegExp>

 #ifdef Q_WS_WIN
 #include <windows.h>
@@ -84,9 +85,11 @@ static bool doOpenUrl(const QUrl& url)
  */
 bool DesktopUtil::openUrl(const QString& url)
 {
-	QByteArray ascii = url.toAscii();
-	if (ascii == url)
-		return doOpenUrl(QUrl::fromEncoded(ascii));
+	// If url only contains characters in the ASCII-range 0x21 (!) - 0x7E (~),
+	// it can be used without further encoding.
+	QRegExp regexp = QRegExp("[\041-\176]*");
+	if (regexp.exactMatch(url))
+		return doOpenUrl(QUrl::fromEncoded(url.toAscii()));
 	else
 		return doOpenUrl(QUrl(url, QUrl::TolerantMode));
 }