It was incorrect to assume there was always a 0th element.  Instead, check that the array is not empty first.

Added new test cases to test empty arrays.

Thanks to David Gobbi for noticing this bug during review of something else.

It's existence seemed to be to support some ancient compilers.

- added range checking to GetAnnotation() for consistency with GetAnnotatedValue()
- improved various comments as I learned the details of these classes
- changed vtkVariant's Valid ivar to bool, the related methods were already using bool
- changed vtkVariant's Type ivar to unsigned int, the related methods were already using unsigned int
- use VTK_VOID instead of a raw 0 for vtkVariant's Type
- make use of constexpr in a few places
- always use .0 so that doubles look like doubles and not ints
- prefer preincrement in for loop
- instead of guarding possible null pointer deref by checking Valid check the pointer itself.

When SPACING and ORIGIN are copied by the pipeline, the associated
DIRECTION should also be copied.  Most VTK filters should simply
pass the DIRECTION to the output if present, a few filters will
have to modify the DIRECTION.

This will increase performances of all code that depends on these functions
such as vtkStringToNumeric, vtkDelimitedTextReader and more.

This tests vtkVariantStringToNumeric

Despite its name, IsItemPresent() doesn't return a boolean, it returns an index.  Made this more clear where it's used.

Introduced a similar method named IndexOfFirstOccurence() that returns the index, as you'd guess from its name.  Changed many uses of IsItemPresent to IndexOfFirstOccurence().

Changed HasProp() and HasImage() from returning int to returning vtkTypeBool. Also, with VTK_LEGACY_REMOVE, changed behaviour to actually return 0 or 1, instead of returning a 1-based index.

This actually comes from `vtkSetGet.h` macros, but these seem to be the
only two files having issues (RogueResearch7's 10.11 builds).

- Mark UseMagnitude in VTK_LEGACY_REMOVE
- HasFixedTopology is totally unused

Determines if the vtkTypeBool typedef is int (backwards compatible) or bool (forward looking).

David Gobbi authored 1 year ago and Mathieu Westphal (Kitware) committed 1 year ago

The file vtkImageData.cxx had control byte 0x0f (^O) at the end
of one of the comment lines.  It has been removed.

Sean McBride authored 1 year ago and Mathieu Westphal (Kitware) committed 1 year ago

Upsized to to int64_t before multiplying.

Fixes some unit test failures under UBSan.

Bengt Rosenberger authored 2 years ago and Mathieu Westphal (Kitware) committed 1 year ago

Wanted to make clear I took that test from TestMappedGridDeepCopy
and didn't entirely write it myself.

Bengt Rosenberger authored 2 years ago and Mathieu Westphal (Kitware) committed 1 year ago

Tests vtkUnstructuredGrid::ShallowCopy with a vtkMappedUnstructuredGrid source
and simulates pipeline execution where executives calls
vtkUnstructuredGrid::Initialize before copy.

Bengt Rosenberger authored 2 years ago and Mathieu Westphal (Kitware) committed 1 year ago

Made vtkUnstructuredGrid::ShallowCopy check and create internal arrays as
necessary if input is a vtkUnstructuredGridBase

Christos Tsolakis authored 1 year ago and Jean-Christophe Fillion-Robin committed 1 year ago

- Remove markdown files
- Exclude vtkModule*cmake since they are now in rst format and rendered in docs.vtk.org
- replace \ref references from headers to the corresponding url in the documentation site.

Although coutnerintuitive, changing the `NewInstance` behavior of
implicit arrays corresponds to more use cases then their current
behavior. The most common use of `NewInstance` in VTK is to create an
empty instance of an object and populate it with some filtered data. In
the case of implicit arrays, this population step is not possible  since
they are read only. Return AOS arrays allows most use cases to just work
out of the box.

Even though things are basically the same, it's best to be explicit with
bridges that do the translations.

Fixes: #18967

Caitlin Ross authored 2 years ago and Ben Boeckel committed 1 year ago

enables changing the precision of loguru timing output.

`vtkAutoInit` is meant to work this way, so always suppress the lint.

Also another near a name

Some classes just need a C++ makeover. Other than that, investigation is
needed (mainly the wrapper tools).

While the lint is generally being suppressed, fix it in headers so that
other projects might have a chance at it being quiet.