As the title says, if right now there is Arch type active (like Level,
Building, etc. etc.), then it's not possible to assign Group to it
automatically (it's being created on root level of the document).
So this patch basically takes an active object and tries to insert it.
Small regression of mine, basically this signal to remove set/locked
state of EditableDatumLabel should be only sent out if current text in
the label is empty or it doesn't contain digits.
Previously it was emitted every intermediate wrong state, so stuff like
"71." was also being matched, and it resulted in resetting the locked
state of the label, which in turn resulted in keeping user from entering
float values.
This patch fixes crashes that we've noticed during migration from Qt5 to
Qt6 in recent months.
If you select items in a tree in a specific direction or range (all, or
from bottom to top) and delete them, there is a high change user will
experience a crash in `testStatus` function. This problem arises because
we're getting into use-after-free situation.
Looking at the callstack there are a lot of calls to
`itemSelectionChanged` during deletion, which takes over item creation
after deletion in `TreeWidget::_slotDeleteObject`. This in turn causes
`DocumentObjectItem::testStatus` to be called prematurely when we have
dangling pointers in object map still. `itemSelectionChanged` signal is
being transmitted because the selection range is changing as we're
constantly deleting and readding certain items.
Previously there was `blockSelection` call during deletion, but it turns
out the signals can still be emitted even AFTER we delete the item. This
had to somehow change between Qt5 and Qt6. So, to be safe, move the
signal block for selection before the obj deletion loop to be sure we
won't retransmit this signal during an uncertain state.
paddlestroke:
> Here we actually need to keep App::Origin. Because we do not allow deletion ONLY of origin objects. Not of normal LCS.
While the original code:
```cpp
auto origin = dynamic_cast<App::Origin*>(lcs);
if (origin && !origin->getInList().empty()) {
```
...handles this perfectly fine, intent isn't obvious when reading it.
Using `is<T>()` shows intent better and should avoid similar situations in the future.
* use early exit to highlight main execution path
* use `auto*` for pointer
* use getObject<T>() instead of casting result from getObject()
* remove empty updateData()
* rename origin to lcs in setTemporaryVisibility for clarity
This is a left-over of the regressions introduced with PR 18126.
Thanks to some moderinization of the code base and replacing static with dynamic casts undefined behaviour
has changed to well-defined behaviour but now unchecked null pointers.
This change does some extra null pointer checks and uses the now correct types for down casting.
Hint: Upstream still uses many static casts here that already cause undefined behaviour when creating a LCS.
This could be the reason for the possible crashes when deleting a LCS as described in 20261
# Conflicts:
# src/Gui/ViewProviderCoordinateSystem.cpp
Until 8de6382 ("Gui: Fix stackoverflow when loading corrupted file")
a static_cast was used to obtain App::Origin object, however with
introducing App::LocalCoordinateSystem this was no longer correct,
although not causing any troubles as OriginFeatures moved into
LocalCoordinateSystem as well.
Recent use of getObject template triggered this problem, so use now
correct cast to App::LocalCoordinateSystem.
Fixes: 19702dc ("Core: Add App::LocalCoordinateSystem")
* Working scale prototype
* Call viewAll to fit geometries in the viewport post-scaling
* Exclude angle dimensions
* Scale the viewport rather than calling viewAll
* Scale dimension annotation along geometries
* Early return when counting more than one dimensional constraint
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Disable sketch autoscale if there are external geometries in the sketch
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Add a setting to disable the feature _ and eventually parametrize
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Check for objects in the viewport in the sketch's ancestry to decide wheter or not to autoscale
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* More consistent camera scaling
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Check for visual indicator in the whole document
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Find visible items in nested assemblies
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Check visual elements in assemblies nested in assemblies
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Set the dimension even if the scaling fails
* Allow constraints that interact with the origin axis/root
* Remove unused variable
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Misc fixes from review
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This adds empty(QSize) method to bitmap factory that creates empty
pixmap. This may seem useless, but after creating bitmap one needs to
remember to clear it and to properly set DPR - BitmapFactory will take
care of it.
Currently if user tries to reset OVP, they can only do that by entering
"0" for example, and then the parameters will get unset in
unsetOnViewParameter. But that will only happen if user types a value
that's under confusion point (typically 1e^-7). In my opinion, it would
be cool to reset that state if user deletes all content in the label, to
allow them to specify coordinates with mouse once again.
Also, this patch fixes a regression with backspace, where deleting stuff
from OVP was working on unix systems, but seems like on Windows it
doesn't pass the check.
