This should fix bug #16974

Code contributed by Jensgw related to bug #16918

The cppcheck warnings were as follows:
style: struct member 'yy_trans_info::yy_verify' is never used.
style: struct member 'yy_trans_info::yy_nxt' is never used.

This commit simplifies the logic checking if builttool target should be
added by adding the variable VTK_COMPILE_TOOLS_IMPORTED.

Since vtkParse has a full c++ preprocessor built in, it is able to
generically expand all macros that it encounters, so the only macros
that are needed are the ones that add wrapping hints.

Some of size_t variables in vtkParse.l were being compared to "int"
variables generated by flex, resulting in compiler warnings. Previously
changes were done to the generated code to fix these warnings, but it
makes more sense to modify vtkParse.l instead.

Types of warnings:
- mismatch between parameters names in declaration vs definition
- make certain parameters references, or const references
- checking array index after its use
- prefer preincrement
- change return type of operator=

There are some very minor API chages here (const/reference).

First, here is some background on CLI response-file processing.
A response-file argument begins with '@', e.g. @/path/to/file.args,
and the first character (the '@') is stripped from the argument to
get the path to the response file.

Due to a serious bug in the wrapper tools, all arguments were checked
to see if they provided response files, regardless of whether the
argument started with '@'.  For example, for the argument '-o', the
the wrapper tools would attempt to read a response file called 'o' and,
since the file didn't exist, it would fall back to correctly processing
the argument as '-o'.

For an arg like '//path/to/file.cxx', however, if '/path/to/file.cxx'
existed then it would be read as a response file, and the wrapper tool
would fail.

These are already informally commented as legacy, this just
allows them to be truly compiled-away.  Will help totally removing them one day,
which is desirable because they violate C++ naming rules by having
2 or more adjacent underscores.

snprintf is now part of C++11 and is already simulated for old versions of VS.

Previously the array was not marked when it was created through
the dataset adapter interface so it had the default FIELD
attribute.

Previously Verdict was used to compute cell sizes but it only
handled triangles, quads, tets and hexes. Adding a new filter
to compute cell sizes and having the volume() method in
algorithms.py to use that instead.

If an out-of-range value is sent to isprint(), then programs generated
by Visual Studio 2010 and later will assert.  I suspect that it does this
because the alternative would be to guess the encoding and potentially
give an incorrect answer, which could compromise the security of the
executable.

Kunda authored 8 years ago and Cory Quammen committed 8 years ago

doxy and misc. typos pt2

Kunda authored 8 years ago and Cory Quammen committed 8 years ago

fixes some doxy typos
fixes misc. comment typos

This is needed to resolve a cycle causing a leak as a result of
add a numpy array to a dataset where the array was derived from
an array associated with the dataset.

Previously, if a class in VTK had no copy constructor it would not be
wrapped (unless it was derived from vtkObjectBase, which always has the
copy constructor disabled).

Examples of classes that are now wrapped are vtkFloatingPointException
(which only has static methods and is never actually instantiated) and
RAII classes like vtkSimpleMutexLock that lack copy constructors for
obvious reasons.

Adds a fallback for python 2.

This is similar to what is already done in ITK and SimpleITK.

The new vtkTargetLinkLibrariesWithDynamicLookup.cmake file is slightly modified copy from
ITK/CMake/itkTargetLinkLibrariesWithDynamicLookup.cmake
The explanation of what this patch tries to achieve is documented in this file.

A new argument is introduced, called OPTIONAL_PYTHON_LINK.
When used, the module will be optionally be linked against libpython.
In the module.cmake files, most vtkPython dependencies were moved to COMPILE_DEPENDS, so that libpython is not added to the target_link_libraries() call.

The vtkPython is explicitely linked against the python libraries, as this is a python executable.

Also, the find_package calls for the PythonLibs were made optional when possible.

The XDMF3 project was not updated, this will need to be done separately if weak linking is wished for that project.

Fixes: #16068

- Avoid cppcheck false positive (due to pointer aliasing confusion). Reworked the code so g doesn’t alias gTmp, so that cppcheck doesn’t think that gTmp is used unititialized.
- added seemingly missing break
- removed store to local variable from a macro
- fixed other warnings.

These are all new warnings vs the previous version, 1.76.1

No real bugs found/fixed here.

Most are either overwriting a never-read variable value,
usually only at intialiazation, or a useless comparison.

In a couple of places, threw in a const or did other minor reworking.

Strictly speaking, in C, unlike C++, one must use void for an empty parameter list.

VTK has several Borland-specific workarounds, but I’m
fairly sure they are all for ancient compilers.
So this changes makes them all apply to older-than-newest
versions only.
This will make it easier to remove them entirely one day,
as currently it’s impossible to know if they are needed for all, old, or new versions.

Apparently expansion of this macro (which is sometimes defined on Win32)
is undefined and causes a compile error.

This is mainly for kwiml, which was added to VTK 7.1.  For type sizes,
endianness, etc. kwiml checks compiler macros instead of using cmake
to generate config headers via try-compile.  Hence, the wrappers must
define all the macros that the native compiler defines.

The parser would die if "volatile" appeared inside curly braces.
This was due to its accidental omission from the "keyword" rule.

Even though typedefs have been handled by the Python wrappers ever since
VTK 6.0, this feature was never enabled for the Java and Tcl wrappers.
This problem was noticed when the vtkMTimeType typedef was introduced,
since it caused MTime methods to disappear from the Java/Tcl wrappers.

ndarray methods such as sum(...) should produce scalar types,
rather than a 0-d VTKArray. Add __array_wrap__(...) method
to coerce 0-d array to scalar.

David Gobbi authored 8 years ago and Chuck Atkins committed 8 years ago

This adds a check to vtkPythonCommand so that it doesn't attempt to
use the calldata pointer if the pointer is NULL.