fix lost error message

[tsteven4]

Why this compiled and output nothing is a mystery to me.

You can test with:
gpsbabel -i kml -f reference/bounds-test.kml -o gpx,gpxver=0.9 -F /dev/null
which should produce a message
GPX: gpx version number "0.9" not valid.

[tsteven4]

This is another way to "fix" it, and a clue to what is wrong:

-    fatal(FatalMsg() << MYNAME ": gpx version number"
-          << gpx_write_version.toString() << "not valid.");
+    qCritical() << MYNAME ": gpx version number"
+          << gpx_write_version << "not valid.";

This results in a message
GPX: gpx version number QVersionNumber(0.9) not valid.

[robertlipe]

LGTM

The best thing about streaming operators is implicit conversions.
The worst thing about streaming operators is implicit conversions.

Related thought. Now that we're c++20, we can rely on std::fmt, no? I'm not sure I want to go nuts replacing printf specifiers everywhere, (in fact, I'm pretty sure I don't...) I resisted it when it meant bringing in a thirdparty (even though that version IS the implementation in most platforms) library, but for debugging, in particular, it can be pretty nice. I'm not sure how well it plays with Qt types[1], but it can't be worse than printf...plus fmt has been pretty widely embraced through the industry so it seems likely that someone has

Compile-time format argument matching testing is pretty nice if we can rely on it.

I'd even give some consideration to exposing these formatting specifiers in our style definitions, but I'm nto really sure how many people are writing their own.

[1] See https://bugreports.qt.io/browse/QTBUG-104651, though that seems mostly internal to Qt itself. Also https://wiki.qt.io/QtCS2024_std::format - notably QTBUG-104652 which dashes my hope above, but looks trivial to both patch in and to disable when our floor hits Qt 6.8 or whatever.

[tsteven4]

This is related to our use of the undocumented operator overload from qversionnumber.h
Q_CORE_EXPORT QDebug operator<<(QDebug, const QVersionNumber &version);
Note the operator operates on and returns a QDebug object, not a QDebug object reference.

If the QVersionNumber was the last thing sent then the code doesn't even compile

/.../gpx.cc: In member function ‘virtual void GpxFormat::wr_init(const QString&)’:
/.../gpx.cc:1003:11: error: cannot bind non-const lvalue reference of type ‘QDebug&’ to an rvalue of type ‘QDebug’
 1002 |     gbFatal(FatalMsg() << "gpx version number"
      |             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1003 |           << gpx_write_version);
      |           ^~~~~~~~~~~~~~~~~~~~
In file included from /.../gpx.h:34,
                 from /.../gpx.cc:22:
/.../defs.h:913:35: note:   initializing argument 1 of ‘void gbFatal(QDebug&)’
  913 | [[noreturn]] void gbFatal(QDebug& msginstance);
      |                           ~~~~~~~~^~~~~~~~~~~
[2/3] Building CXX object CMakeFiles/gpsbabel.dir/tpo.cc.o

[tsteven4]

I'm not sure I want to go nuts replacing printf specifiers everywhere, (in fact, I'm pretty sure I don't...)

Me either!