That bug does not explain to me why we can't use unsigned short columns. 

This might have been fixed for gtkmm 2.4, as part of this bug:
http://bugzilla.gnome.org/show_bug.cgi?id=102853

I don't understand. The problem described in
http://bugzilla.gnome.org/show_bug.cgi?id=102853
seems to impose that compile time errors arise when using unsuppoerted
types for TreeView::append_column_editable(). I'm talking about
TreeView::append:column(), however.

While
http://bugzilla.gnome.org/show_bug.cgi?id=102853
is certainly somehow related to
http://bugzilla.gnome.org/show_bug.cgi?id=94170
I can't see how it is supposed to be fixed with the patch
http://bugzilla.gnome.org/showattachment.cgi?attach_id=13428

Since I don't have gtkmm-2.4 installed (which would take a while on
this damn slow machine) I can't check myself whether the problems are
gone in 2.4 so I'm attaching a test case that uses
TreeModelColumn<unsigned short> and results in the following run time
errors with gtkmm-2.2:

(process:1270): GLib-GObject-CRITICAL **: gtype.c:1875: initialization
assertion failed, use g_type_init() prior to this function
 
(a.out:1270): Gtk-WARNING **: gtkliststore.c:377: Invalid type (null)
passed to gtk_list_store_set_column_types
 
 
(a.out:1270): GLib-GObject-CRITICAL **: file gvalue.c: line 302
(g_value_type_transformable): assertion `G_TYPE_IS_VALUE (src_type)'
failed
 
(a.out:1270): GLib-GObject-WARNING **: gtype.c:2967: type id `0' is
invalid
 
(a.out:1270): GLib-GObject-WARNING **: can't peek value table for type
`<invalid>' which is not currently referenced
 
(a.out:1270): GLib-GObject-WARNING **: gvalue.c:93: cannot initialize
GValue with type `(null)', this type has no GTypeValueTable implementation
 
[2]+  Segmentation fault      (core dumped) ./a.out


(Note, however, that I get different runtime errors and no segfault
with my app which loads the treeview from a .glade file with
libglademm - strange!)

Please try the test case with gtkmm-2.4 and tell me what the correct
behaviour should be. I don't think we can ignore this - it really took
me a while to locate the error in my app.

Created attachment 18867 [details]
test_treeview_unsigned_short.cc

Yes, I looked at that bug again and it does seem to solve a different
problem. 

Tried it out with gtkmm 2.2.5, and it obviously has not been fixed.
Here's the output, in case anyone is interested:

 (process:16668): GLib-GObject-CRITICAL **: gtype.c:1875: 
initialization assertion failed, use g_type_init() prior to this 
function

(a.out:16668): Gtk-WARNING **: gtkliststore.c:377: Invalid type 
(null) passed to gtk_list_store_set_column_types


(a.out:16668): GLib-GObject-CRITICAL **: file gvalue.c: line 302 
(g_value_type_transformable): assertion `G_TYPE_IS_VALUE (src_type)' 
failed

(a.out:16668): GLib-GObject-WARNING **: gtype.c:2967: type id `0' is 
invalid

(a.out:16668): GLib-GObject-WARNING **: can't peek value table for 
type `<invalid>' which is not currently referenced

(a.out:16668): GLib-GObject-WARNING **: gvalue.c:93: cannot 
initialize GValue with type `(null)', this type has no 
GTypeValueTable implementation

Segmentation fault (core dumped)

I will investigate eventually if nobody else does.

I couldn't manage half the things I wanted to tackle this weekend :(
However, I will drop a note before I start investigations of my own
before you do. Might get Friday, though. You are sure now, that there
were no fixes for gtkmm-2.4 regarding this problem?

No, but it looks like Daniel had some clue. The hard drive on my linux
machine will be kaput until Monday.

I will start to investigate now.
Questions to be answered:
- Is this a bug?
- Otherwise: Why not? Can error messages be triggered at compile time?
- Otherwise: Where shall we document this behaviour?

Now I know how to reproduce the other error message and avoid the
segfault. I'll attach a corrected test case. Difference: don't
instantiate a Gtk::TreeModelColumnRecord before calling "Gtk::Main
kit;". Otherwise Glib::Value<unsigned short>::value_type() fails
triggering the first run-time error and eventually leading to the
segfault.

The corrected test case triggers the following run-time error known
from http://bugzilla.gnome.org/show_bug.cgi?id=94170:

(a.out:1163): GLib-GObject-WARNING **: unable to set property `text'
of type `gchararray' from value of type `glibmm__CustomBoxed_t'

Created attachment 19038 [details]
test_treeview_unsigned_short_2.cc

> - Is this a bug?

First test case: no: use of gtk(mm) functions (here for setting up
custom GValue type) before instantiating Gtk::Main is not allowed.

Second test case: not really: use of custom GValue types is perfectly
legal. However it is not possible to automatically create a
TreeViewColumn from TreeView::append_column(). Instead, the
TreeViewColumn has to be created manually and a function for
displaying data has to be provided with
TreeViewColumn::set_cell_data_func(). After that, the column can be
appended directly with the appropriate TreeView::append_column() overload.


> - Otherwise: Why not? Can error messages be triggered at compile time?

First test case: No.

Second test case: In theory. The TreeView::append_column(const
Glib::ustring& title, const TreeModelColumn<ColumnType>& model_column)
overload should fail if the TreeModelColumn uses a custom GValue type.
A compile-time-check can be added using some template magic with
template specializations for all supported types. The corresponding
TreeView::append_column_editable() overload already fails to compile
if the type is not supported. A run-time-check might be easier and
cleaner to implement (somehow check Value<the_type>::value_type()).


> - Otherwise: Where shall we document this behaviour?

First test case: Already documented in
file:///usr/local/share/doc/gtkmm-2.0/docs/reference/html/classGtk_1_1TreeModelColumnRecord.html

Second test case:
file:///usr/local/share/doc/gtkmm-2.0/docs/tutorial/html/ch08s02.html#id2962983
and
file:///usr/local/share/doc/gtkmm-2.0/docs/reference/html/classGtk_1_1TreeView.html#a25
should mention that TreeViewModelColumns that use custom GValue types
can not be used this way. The standard GValue types have to be listed
somewhere.

> > - Otherwise: Why not? Can error messages be triggered at compile time?
> 
> First test case: No.
> 
> Second test case: In theory. The TreeView::append_column(const
> Glib::ustring& title, const TreeModelColumn<ColumnType>& model_column)
> overload should fail if the TreeModelColumn uses a custom GValue type.

I was wrong here. It is also possible to set the cell data function
with TreeViewColumn::set_cell_data_func() after the column has been
created with the TreeView::append_column(const
Glib::ustring& title, const TreeModelColumn<ColumnType>& model_column)
overload. The TreeViewColumn doesn't have to be created manually.
=> We only need to improve the documentation. I will attach a patch.

Created attachment 19059 [details] [review]
gtkmm2-improve_treeview_docs.diff: Proposed documentation additions.

The only problem I see after improving the documentation of
TreeView::append_column is that I find it quite difficult to track
down the runtime error

(a.out:1163): GLib-GObject-WARNING **: unable to set property `text'
of type `gchararray' from value of type `glibmm__CustomBoxed_t'

to an incorrect use of this function. Maybe we should cite the error
message?

You might also try doing the same thing in GTK+ in C.

Please commit the documentation patch after making these changes:

> Note that only for some basic column types like Glib::ustring, int,
bool, Gdk::Pixbuf, etc. the CellRenderer is set up correctly. ...

I would prefer
"
The CellRenderer can only be created automatically for some basic
column types, such as Glib::ustring, int, bool, Gdk::Pixbuf. Also, the
default CellRenderer might not be setup exactly as needed.

You might prefer to create the TreeViewColumn and or CellRenderer
manually. You might provide a callback that converts the type into a
string representation with TreeViewColumn::set_cell_data_func().
"

Please do not mention this:
> * If you want this function to support a custom CellRenderer provide
a template specialization of
TreeViewColumn_CellRendererGeneration::generate_cellrenderer()
That is fairly internal API and this is an unpleasant thing for
someone to do.

I will do so this evening. I must not forget to also commit to gtkmm-
2.4!

> You might also try doing the same thing in GTK+ in C.
In C you also have to provide a callback with 
gtk_tree_view_column_set_cell_data_func() for unsupported types such 
as (unsigned) short. There is no knowledge to be gained with a C test 
case so I won't waste my time with it...

> I would prefer
> "
> The CellRenderer can only be created automatically for some basic
> column types, such as Glib::ustring, int, bool, Gdk::Pixbuf. Also,
> the default CellRenderer might not be setup exactly as needed.
> You might prefer to create the TreeViewColumn and or CellRenderer
> manually. You might provide a callback that converts the type into a
> string representation with TreeViewColumn::set_cell_data_func().
> "

Please also proof-read the following paragraph I'd like to add:

 * Otherwise, if the type is not supported, at run-time no text will
 * appear in the column while the following warning is generated
repeatedly:
 * GLib-GObject-WARNING **: unable to set property `text' of type
 * `gchararray' from value of type `glibmm__CustomBoxed_t'

I will wait with the commit. Meanwhiles, I'm attaching updated patches
for gtkmm-2.2 and gtkmm-2.4.

Created attachment 19126 [details] [review]
gtkmm2-improve_treeview_docs.diff: updated patch for gtkmm-2.2

Created attachment 19128 [details] [review]
gtkmm-improve_treeview_docs.diff: updated patch for gtkmm-2.4

OK, though we would of course prefer to have a more meaningful error 
generated from gtkmm.

I need to investigate why the unsigned type is not supported by GTK+.

Ok, then I will commit as soon as possible.

> I need to investigate why the unsigned type is not supported by 
GTK+.
It's not the unsigned but the short type. int and unsigned int are 
supported, short and unsigned short are not. There is no GValue type 
for them.

I guess this is platform-dependent.

Documentation additions committed to both gtkmm2 and gtkmm cvs modules.

This might be a clue about how to support these extra types:
http://bugzilla.gnome.org/show_bug.cgi?id=83190

This bug seems relevant too:
http://bugzilla.gnome.org/show_bug.cgi?id=94170

I think we just need to add an unsigned short template specialization. I will
try it.

Created attachment 33892 [details]
test_treeview_short.cc

Here's a slightly simpler example, updated for gtkmm 2.4.

I must find out how the CellRendererText is able even to render an int via it's
text property.

I did some reasearch, and I have a possible solution:

We could use g_value_register_transform_func() to register a new transformation
from the Value<short>::value_type() to gchararray. glib already registers a
transformation func for int to gchararray. We would do this in our
Value<>::value_type() specialization.

Alternatively, we could persuade the glib developers to do it in glib.

However, all of these number-to-string conversions are arbitrary and probably
not often what is needed, so people often use set_cell_data_func(). Maybe an
append_column_numeric(model_column, numeric_format) would be better, if we can
think of a sensible way to represent the format.

I have added TreeView::append_column_numeric() and
TreeView::append_column_numeric_editable() in gtkmm 2.5, which can at least be
used as workarounds.

Hi,

I post on this bug because I've got same or a related problem (after many years as I can see in the dates of comments).

I had the same problem with short types; seeing this bug, I've used append_column_numeric like Murray said and it works, but using append_column_numeric_editable, the code doesn't compile, throwing this message:

In file included from /usr/include/gtkmm-2.4/gtkmm/combobox.h:34:0,
                 from /usr/include/gtkmm-2.4/gtkmm.h:123,
                 from main.cc:3:
/usr/include/gtkmm-2.4/gtkmm/treeview.h: In function ‘void Gtk::TreeView_Private::_auto_store_on_cellrenderer_text_edited_string(const Glib::ustring&, const Glib::ustring&, int, const Glib::RefPtr<Gtk::TreeModel>&) [with ColumnType = short int]’:
/usr/include/gtkmm-2.4/gtkmm/treeview.h:2463:82:   instantiated from ‘void Gtk::TreeView_Private::_connect_auto_store_editable_signal_handler(Gtk::TreeView*, Gtk::CellRenderer*, const Gtk::TreeModelColumn<ColumnType>&) [with ColumnType = short int]’
/usr/include/gtkmm-2.4/gtkmm/treeview.h:2181:5:   instantiated from ‘int Gtk::TreeView::append_column_numeric_editable(const Glib::ustring&, const Gtk::TreeModelColumn<ColumnType>&, const Glib::ustring&) [with ColumnType = short int]’
main.cc:63:57:   instantiated from here
/usr/include/gtkmm-2.4/gtkmm/treeview.h:2492:9: error: invalid cast from type ‘const Glib::ustring’ to type ‘short int’

And, for 3.0:

In file included from /usr/include/gtkmm-3.0/gtkmm/combobox.h:36:0,
                 from /usr/include/gtkmm-3.0/gtkmm/appchooserbutton.h:27,
                 from /usr/include/gtkmm-3.0/gtkmm.h:101,
                 from main.cc:3:
/usr/include/gtkmm-3.0/gtkmm/treeview.h: In function ‘void Gtk::TreeView_Private::_auto_store_on_cellrenderer_text_edited_string(const Glib::ustring&, const Glib::ustring&, int, const Glib::RefPtr<Gtk::TreeModel>&) [with ColumnType = short int]’:
/usr/include/gtkmm-3.0/gtkmm/treeview.h:2242:82:   instantiated from ‘void Gtk::TreeView_Private::_connect_auto_store_editable_signal_handler(Gtk::TreeView*, Gtk::CellRenderer*, const Gtk::TreeModelColumn<ColumnType>&) [with ColumnType = short int]’
/usr/include/gtkmm-3.0/gtkmm/treeview.h:1960:5:   instantiated from ‘int Gtk::TreeView::append_column_numeric_editable(const Glib::ustring&, const Gtk::TreeModelColumn<ColumnType>&, const Glib::ustring&) [with ColumnType = short int]’
main.cc:63:57:   instantiated from here
/usr/include/gtkmm-3.0/gtkmm/treeview.h:2271:9: error: invalid cast from type ‘const Glib::ustring’ to type ‘short int’

Using integer and casting resolves the problem, but this is a workaround.

I don't know if this deserves to be solved or not, but I notify it to put record.

I've tried with gtkmm 2.4 (apt: 1:2.24.0-1) and 3.0 (3.0.1-1), on Debian testing.

> using
append_column_numeric_editable, the code doesn't compile

Please file a new bug, ideally with a simple-as-possible test case.

(In reply to comment #35)
> Please file a new bug, ideally with a simple-as-possible test case.

Done: https://bugzilla.gnome.org/show_bug.cgi?id=655416