There is a possible memory leak in case the SWIG_exception_fail macro
is called. The problem is related to its definition that call the
function Rf_warning. This function (as well as Rf_error) involves
a longjmp over C++ destructors on the stack. Thus, all the objects
allocated on the heap are not freed.
Closes#914
There is some consolidation work to be done. The
core of the change is getRClassName2, which will
probably eventually replace getRClassName.
getRClassName seems to be in a funny state, with
a middle argument that is commented out and never
used.
My next step is to verify whether the new version
can replace it.
arbitarily complex value expressions.
R-side initialization of the enumeration structure employs
delayedAssign, which allows the initialization code to be
sourced before loading the shared library, which is the
usual behaviour for code in an R package.
The -cppcast option is still turned on by default. The -nocppcast option
to turn off the use of c++ casts (const_cast, static_cast etc) has been
removed. However, defining SWIG_NO_CPLUSPLUS_CAST will still generate C casts
instead of C++ casts for C++ wrappers.
This a revert of commit fc79264a48:
"Revert "Remove -cppcast and -nocppcast command line options""
The Scilab and Javascript casting problems are now fixed, so -cppcast
is now switched on as default.
The -cppcast option is still turned on by default. The -nocppcast option
to turn off the use of c++ casts (const_cast, static_cast etc) has been
removed. However, defining SWIG_NO_CPLUSPLUS_CAST will still generate C casts
instead of C++ casts for C++ wrappers.
Replace R_SWIG_EXTERNAL and R_SWIG_OWNER with the standard equivalent
macros. I can't see where these were actually used, but in case they are,
equivalent backwards compatibility can be obtained using:
#define R_SWIG_EXTERNAL 0
#define R_SWIG_OWNER SWIG_POINTER_OWN
SWIG_MakePtr does not seem to be used within the R SWIG code base so I
havn't added the above as backwards compatibility macros.
There seems to be no memory management in R. I've made this change
to fit in with the rest of SWIG though in preparation for an upcoming
commit to use the same shared_ptr macros as is used elsewhere.
The function "OutputClassMemberTable" tags methods generated
for a class so that R wrapper code supporting syntax like:
WrappedClass$VarName <- value
The comments in the function indicate that it is looking for
method names ending in "_set", but in fact it was searching for
methods ending in just "set", This was resulting in class methods
ending in set (names like GetOffset, SetOffset etc) being placed
in the wrong accessor functions and thus not being available in
the normal way.
There is probably a more appropriate way to tag the _set functions
when they are being created, which is a future step. This
patch makes the behaviour conform to the original intention.
The function "OutputClassMemberTable" tags methods generated
for a class so that R wrapper code supporting syntax like:
WrappedClass$VarName <- value
The comments in the function indicate that it is looking for
method names ending in "_set", but in fact it was searching for
methods ending in just "set", This was resulting in class methods
ending in set (names like GetOffset, SetOffset etc) being placed
in the wrong accessor functions and thus not being available in
the normal way.
There is probably a more appropriate way to tag the _set functions
when they are being created, which is a future step. This
patch makes the behaviour conform to the original intention.
includes the addition of a _runme for an existing test - preproc_constants
that was previously not run. That tests includes a preprocessor based
setting of an enumeration which is ignored by the existing r enumeration
infrastructure. The new version correctly reports the enumeration value
as 4 - previous versions set it to 0. Traditional enumerations are unchanged.
The approach used to deal with these enumerations is similar to that of
other languages, and requires a call to a C function at runtime to return
the enumeration value. The previous approach figured out the values statically
and this is still used where possible. The need for a runtime call leads to
changes in when swig code is used in packages - see below.
One test that previously passed now fails - namely the R sourcing of
preproc_constants.R, as the enumeration code requires the shared library,
which isn't loaded by that script.
There is also a modification to the way the R _runme.R files are used.
The call to R CMD BATCH now includes a --args option that indicates
the source folder for the unittest.R file, and the first couple
of lines of the _runme.R files deal with correctly locating this.
Out of source tests now run correctly.
This work was motivated by problems generating the SimpleITK binding,
specifically with some of the more complex enumerations.
This approach does have some issues wrt to code in packages, but I can't
see an alternative. The problem with packages is that the R code setting
up the enumeration structures requires the shared library so that the C
functions returning enumeration values can be called. The enumeration
setup code thus needs to be moved to the package initialisation section.
For SimpleITK I do this using an R script, which I think is an acceptable
solution. The core part of the process is the following function. I dump
all the enumeration stuff into a .onload function. This is only necessary
if some of the enumerations are tricky.
splitSwigFile <- function(filename, onloadfile, mainfile)
{
p1 <- parse(file=filename)
getdefineEnum <- function(X)
{
return (is.call(X) & (X[[1]]=="defineEnumeration"))
}
dd <- sapply(p1, getdefineEnum)
enums <- p1[dd]
enums <- unlist(lapply(enums, deparse))
enums <- c(".onLoad <- function(libname, pkgname) {", enums, "}")
everythingelse <- p1[!dd]
everythingelse <- unlist(lapply(everythingelse, deparse))
writeLines(everythingelse, mainfile)
writeLines(enums, onloadfile)
}
