https://bugs.documentfoundation.org/show_bug.cgi?id=173809

--- Comment #15 from Michael Weghorn <[email protected]> ---
(In reply to BogdanB from comment #12)
> Michael, please, can you take a look?

>From a first analysis: The treeview seems to try to use both, the "children on
demand" feature (which adds a dummy entry for the gtk and vcl implementations)
and "real"/"actual" children for *the same entry* at the same time.

Seeing that the Navigator apparently tries to use both at the same time is very
unexpected and surprising to me. 

The new implementation from the commit mentioned in comment 11 assumes that
*either* the "children on demand" feature is enabled and there are no actual
child entries to consider or considers the actual children of an entry.

After re-reading the "expanding on-demand node details" documentation (in
include/vcl/weld/TreeView.hxx), this still seems weird/unexpected to me, so I
wonder whether what Navigator does here is actually intended to work the way it
did in the past or it should rather always only have either the actual nodes or
the "children-on-demand" flag set for its entries.

(My gut feeling is that a fix might best be done in Navigator, not the
weld::TreeView implementation to somehow retrofit the previous behavior in
there.)

Demo Gerrit change that demonstrates the difference between the implementations
and prints the current children of the node where they differ:
https://gerrit.libreoffice.org/c/core/+/213077

>From the sample output also mentioned in the Gerrit change:

    warn:vcl:2053203:2053203:vcl/unx/gtk3/gtkinst.cxx:15616: child count: 3
    warn:vcl:2053203:2053203:vcl/unx/gtk3/gtkinst.cxx:15621: child 0: <dummy>
    warn:vcl:2053203:2053203:vcl/unx/gtk3/gtkinst.cxx:15621: child 1: 1 First
subtitle
    warn:vcl:2053203:2053203:vcl/unx/gtk3/gtkinst.cxx:15621: child 2: 2 Second
subchapter


@Jim: Any thoughts?

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to