Hello,
I found two GDB Debugger watches related bugs in revision 3515:
1) there are two dialogs for editing watches format -- EditWatchesDlg [Menu->Debug->Edit watches] and EditWatchDlg [accessible
through debug window Watches, rightclick and "Edit watch"]. On the latter there are only 8 formatting options, the "Floating point" is missing.
It can be fixed by adding line in edit_watches_dlg.xrc, here is svn diff
Index: edit_watches_dlg.xrc
===================================================================
--- edit_watches_dlg.xrc (revision 3515)
+++ edit_watches_dlg.xrc (working copy)
@@ -201,6 +201,7 @@
<item>Hexadecimal</item>
<item>Binary</item>
<item>Character</item>
+ <item>Floating point</item>
</content>
<selection>0</selection>
<dimension>2</dimension>
2) quite a "big" bug - when refreshing the tree with watches [debuggertree.cpp, method BuildTree on line 154], you create a new list with new items,
compare them to the items in the tree, modify the existing [color red when a change is found] and you DELETE the items in the tree that now have no value.
But it is done in a wrong way. To see this, watch an array with 10 values and then set the arraysize to 8 values. Then the array is displayed using singleline mode. But in the tree, 4 children remains! [in next refresh, only 1 and then all chindren are gone]. Let's have a look at the skeleton of the BuildTree method:
void DebuggerTree::BuildTree(WatchTreeEntry& entry, wxTreeItemId parent)
{
...
wxTreeItemId item = m_pTree->GetFirstChild(parent, cookie);
size_t count = 0;
while (item) {
// if we still have entries
if (count < entry.entries.size())
{ ... }
else {
// no more entries; delete item
wxTreeItemId newitem = m_pTree->GetNextChild(parent, cookie);
m_pTree->Delete(item);
item = newitem;
}
}
...
}
The GetNextChild function uses cookie to store "the last child I returned from GetFirst/NextChild call". I suggest it uses the index of last child. But when you do m_pTree->Delete(item) BETWEEN the GetNextChild calls, then the cookie is invalidated -- if they use an index, next GetNextChild call SKIPS one child!
[I hope I made myself clear. The point is, you cannot delete children of a node between GetNextChild calls].
I couldnt find better solution than storing the children in a stl_vector and then deleting them at once. Here is my patch using svn diff, if you would like it:
Index: debuggertree.cpp
===================================================================
--- debuggertree.cpp (revision 3521)
+++ debuggertree.cpp (working copy)
@@ -167,6 +167,7 @@
#endif
wxTreeItemId item = m_pTree->GetFirstChild(parent, cookie);
size_t count = 0;
+ std::vector<wxTreeItemId> to_delete;
while (item)
{
// if we still have entries
@@ -193,10 +194,14 @@
{
// no more entries; delete item
wxTreeItemId newitem = m_pTree->GetNextChild(parent, cookie);
- m_pTree->Delete(item);
+ to_delete.push_back(item);
item = newitem;
}
}
+ while (to_delete.size()) {
+ m_pTree->Delete(to_delete.back());
+ to_delete.pop_back();
+ }
// now add any remaining entries
for (; count < entry.entries.size(); ++count)
when I found a bug, i like to find a solution as well. Should I send these svn diffs along, or does it bothers you?
Have a nice day,
Milan Straka
