I can second this request, and I think it's rather a severe bug than just a pure
wish.

At least mozilla and opera convert the accents that do not fit into the charset
of the page containing the form to unicode entities, so e.g. if the page is
iso-8859-1 then an euro-sign becomes "€" (which, in case of a GET submit
type, gets even further escaped to %26%238364%3B in the URL).

This way no information is lost, the user will most likely see exactly the same
character he entered on the resulting page (unless the server explicitely does
something for things to go bad).

The current konqueror behavior leads to data loss without notifying the user,
and furthermore assumes some character set knowledge from the user, which it
shouldn't, users shouldn't need to know what iso-8859-1 or unicode or utf-8 are,
they should only see things working properly.

Sorry, forgot to say, I have KDE 3.3.

Wow, someone saw that bug report at last!
I hope I can point the KDE developers' attention to it, so I we won't have to wait much longer for a fix.

Created an attachment (id=7837) [details]
Patch for ampersand-escaping characters

The attached patch should be applied to khtml/html. I have not fully tested it,
but try and see if it helps.

Works nice for me, thanks.

Just a small question... I'm not familiar with the Unicode handling of Qt, but
.unicode() returning an unsigned short really shocked me, as Unicode definitely
has characters above 65536. Is there anything known how this will be handled
in future versions of Qt? Will a new function be introduced, or the return
value of .unicode() extended to unsigned long?

In the latter case, I'd recommend to use
  ampersandEscape.sprintf("&#%lu;", (unsigned long)(c.unicode()));
instead of
  ampersandEscape.sprintf("&#%hu;", c.unicode());
it doesn't hurt, but works better if Qt changes to ucs4.

CVS commit by carewolf: 

Ampersand-escape otherwise unencodable characters. 
Matches Gecko behavior.
FEATURE:82018


  M +6 -2      ChangeLog   1.303
  M +25 -7     html/html_formimpl.cpp   1.387 [POSSIBLY UNSAFE: printf]


--- kdelibs/khtml/ChangeLog  #1.302:1.303
@@ -1,2 +1,6 @@
+2004-10-16  Allan Sandfeld Jensen <kde@carewolf.com>
+        * html/html_formimpl.cpp: Escape otherwise unencodable characters.
+        Matches the behavior of Gecko.
+
 2004-10-15  Stephan Kulow  <coolo@kde.org>
 
@@ -4,5 +8,5 @@
         got items when we calculate a height for items (#87466)
 
-        * css/html4.css: changing default horizontal margins for H1-H6 from 
+        * css/html4.css: changing default horizontal margins for H1-H6 from
         auto to 0 (#91327)
 
@@ -33,5 +37,5 @@
         * rendering/render_block.cpp (layoutBlockChildren):
         simpler implementation for compact display: do not insert the
-        compact child within the next block anymore. 
+        compact child within the next block anymore.
         Solves lot of problems with host blocks having non-inline children.
 

--- kdelibs/khtml/html/html_formimpl.cpp  #1.386:1.387
@@ -174,7 +174,25 @@ static QCString encodeCString(const QCSt
 }
 
+// ### This function only encodes to numeric ampersand escapes,
+// ### we could use standard ampersand values as well.
+inline static QString escapeUnencodeable(const QTextCodec* codec, const QString& s) {
+    QString enc_string;
+    int len = s.length();
+    for(int i=0; i <len; i++) {
+        QChar c = s[i];
+        if (codec->canEncode(c))
+            enc_string.append(c);
+        else {
+            QString ampersandEscape;
+            ampersandEscape.sprintf("&#%u;", c.unicode());
+            enc_string.append(ampersandEscape);
+        }
+    }
+    return enc_string;
+}
+
 inline static QCString fixUpfromUnicode(const QTextCodec* codec, const QString& s)
 {
-    QCString str = codec->fromUnicode(s);
+    QCString str = codec->fromUnicode(escapeUnencodeable(codec,s));
     str.truncate(str.length());
     return str;

Thanks a lot for the fix!
I'll try it on the next KDE upgrade.