===============================================
On Undo/Redo there should not be an unconditional recompute.
The need for this recompute is probably related to the fact that PropertyConstraintList
has invalid geometry keys.
=======================================================================
On App:Document::undo, applchn is called on every transactional object (e.g. properties) affected, which
calls Paste on the property.
Each of the properties of SketchObject cause a call to SketchObject::OnChanged, and App::DocumentObject::OnChanged, the latter
calling onChangeProperty, which signals the viewprovider from Gui::Document::slotChangedObject via ViewProviderSketch::updateData.
This causes that the ViewProvider is updated when Constraints and Geometry indexes are not matching.
The solution proposed has three parts:
1. First, at ViewProvider::updateData update is prevented while undo/redo transaction is performed by checking isPerformingTransaction()
2. Second, the onUndoRedoFinished() mechanism of SketchObject causes the call of updateData when it solves the sketch (and sets the solved geometry)
3. Third, Gui::Document::signalUndoDocument and Gui::Document::signalRedoDocument (via the slots in ViewProviderSketch) are
used to perform the recompute of the ViewProvider when the undo/redo transaction is finished if needed.
this is unnecessary since Qt will automatically use the available font. It only raises questions since one see in the commit diffs that depending on who modified the dialog, his system default font will appear. (like in the initial commit of this PR)
- I had a look at Crowdin and saw that the translators don't know what to do with the uppercase letter words. And in fact we can use "normal" terms
- I also fixed a typo (missing 'd') to be consistent
'QString& QString::vsprintf(const char*, __va_list_tag*)' is deprecated: Use vasprintf(), arg() or QTextStream instead [-Wdeprecated-declarations]
'QString& QString::sprintf(const char*, ...)' is deprecated: Use asprintf(), arg() or QTextStream instead [-Wdeprecated-declarations]
Since Py3.3: 'Py_ssize_t PyUnicode_GetSize(PyObject*)' is deprecated [-Wdeprecated-declarations]
Since Py3.9: 'PyObject* PyEval_CallObjectWithKeywords(PyObject*, PyObject*, PyObject*)' is deprecated [-Wdeprecated-declarations]
Since Py3.9: 'void PyEval_InitThreads()' is deprecated [-Wdeprecated-declarations]
- everything that is part of the drawing should be plain SVG (strictly following the SVG specification) to assure that every SVG program can handle it (e.g. Internet browsers).
The hatch patterns however contained old stuff like traces of Sodipodi that might be a problem in the future.
- negative scale and line width is not sensible and lead to strange effects thus set a minimum
- set the minimum reasonably above 0 since e.g. a hatch scale of 0.01 fills 8 GB RAM -> out of RAM error
- disable KeyboardTracking since we don't want a time-consuming recomputation while the user changes a value
The use of `App::PropertyLinkSubList` for `'PathSubelements'`
is a mistake because we could use a single `App::PropertyLinkSub`
to handle both `'PathObject'` and `'PathSubelements'` properties.
This commit doesn't change any code, it just adds comments
explaining the situation so that it is not forgotten,
and we remember to address it in the future.
Ideally we should migrate the objects, but we may also decide
to break compatibility with older `PathArrays` if both properties
can't be migrated easily.
