- Since (at least) Octave 5.1.0, the Octave error() function now raises a C++
exception, which if uncaught immediately exits a SWIG wrapper function,
bypassing any cleanup code that may appear after a "fail:" label.
- This patch adds a "try { ... } catch(...) { }" block around the contents of
SWIG wrapper functions to first execute the cleanup code before rethrowing any
exception raised.
- It is backward compatible with earlier versions of Octave where error() does
not raise an exception, which will still branch to the "fail:" block to
execute cleanup code if an error is encountered.
- This reverts commit 931656bcbe
- Since atexit() does not pass along the desired exit status,
__swig_atexit__() always exits with status zero, regardless
of whether Octave completed successfully or raised an error.
- This means the success/failure of Octave scripts which load
SWIG modules cannot be determined by other programs, which
makes them unusable.
- Instead, provide a Octave function swig_exit() which calls
::_Exit() with a given exit status. This way at least a
clean exit from Octave can be guaranteed for future versions
if the seg-fault problem is not fixed.
- The {DECLARE|DEFINE}_OV_TYPEID_FUNCTIONS_AND_DATA declarations
attached to this class cause a seg-fault in the module_load
Octave example. Removing these declarations fixes the seg-fault.
- While cause of seg-fault is unknown, note that (in Octave 5.1.0)
the declaration of octave_function, which is the base class for
octave_swig_bound_func, does not use these declarations. So it's
possible they simply are not required for this type of subclass.
- As of Octave 5.1.0, mkoctfile no longer works with 'env -'
(no environment), so need to pass at least PATH and
LD_LIBRARY_PATH for it to work.
- Still need to clear environment so that it doesn't override
mkoctfile defined variables, e.g. CXXFLAGS.
Compiling code including Guile headers with default compiler options
doesn't work any more since a recent (~2020-05-05) update to Guile 3.0.2
on Homebrew (2.2.7 was used previously) due to
error: redefinition of typedef 'scm_print_state' is a C11 feature
in libguile/print.h and scm.h headers.
Work around this by enabling C11 for this build by explicitly setting
CSTD. Note that we can't just use CPP11=1 for this build, because this
would also enable C++11-specific tests that are not currently supported
by Guile module.
This error is given for any use of variable called "l" (and also "I" and
"O", but we don't seem to have any of those) and it doesn't seem to be
worth changing this variable name in the tests code, as it's really not
that meaningful there anyhow, so just disable the warning to let the CI
builds, which now use pycodestyle 2.6, instead of 2.5 which didn't have
this error and which is still the latest available in Debian Sid, pass.
No real changes, just use simpler quoting construct which is sufficient
(as we don't need to expand any variables) and also happens not to break
Vim 8 syntax highlighting in the entire file, unlike double quotes.
The "endlink" command is processed in processWordCommands, which by
default skips space occuring after the command, which is intended for
removing leading space from a command argument. For "end" commands,
we don't want to do this. Note that certain end commands such as
"endcode" aren't processed by processWordCommands (believe
addCommandUnique ends up handling them).
Update usage of \link in doxygen_translate_all_tags.i to test handling
of space after \endlink.
Tweaking some of the usage in doxygen_misc_constructs.i to remove what
seems to be an extra space from the input (otherwise we would need to
add an extra space to the expected output).
This fixes EOF detection on platforms where char is unsigned, as
comparing it with EOF could never return true there.
Thanks gcc for the warning "comparison is always true due to limited
range of data type [-Wtype-limits]".
* abstract-conversion-operators:
Add test cases for abstract user-defined conversion operators
Recognize C++ conversion operators with trailing '= 0' as abstract
When wrapping STL containers, remove a shared_ptr reference count
increment when an upcast is needed when checking type conversion
in traits_check::check.
This test code tests the upcast code:
swig_assert_equal_simple(-1, base_num2([Derived(7)]))
Although there is no explicit memory leak fix test, it does at least run the code.
Handling of None needs fixing in Python (it is working in Ruby)
Note that the Ruby implementation has a partial template specialization
for shared_ptr, whereas the Python implementation is in the generic
code!
Issue #1512
* master-fix-vector-shared_ptr:
fixing memleak of shared_ptr objects in python with creating a [wrapped] vector<shared_ptr<Foo>> from a list of shared_ptr<Bar> where Bar is derived from Foo.
If python_append.i is modified to use:
__slots__ = []
instead of
__slots__ = ["this"]
then this additional error checking prevents a crash and shows a stack trace and error:
AttributeError: 'ForSlots' object has no attribute 'this'
Related to issue #1674
