[Rcppoctave-user] RcppOctave on Windows: testing needed

[Dominick Samperi]

BTW, in case this is not obvious, the function in strtest.cpp is
not called at runtime! Thus the only way it could lead to problems
is if the classes (like string_vector) need startup (static) initialization.


On Thu, Oct 10, 2013 at 9:45 AM, Dominick Samperi <djsamperi at gmail.com>wrote:

> Your modified strtest.cpp doesn't help:
>
> ** R
> ** inst
> ** preparing package for lazy loading
> ** help
> *** installing help indices
> ** building package indices
> ** installing vignettes
> ** testing if installed package can be loaded
> sh: line 1: 28714 Abort trap: 6
> '/Library/Frameworks/R.framework/Resources/bin/R' --no-save --slave 2>&1 <
> '/var/folders/fq/2th4vc9n1mq9cxv5nl9bmzkc0000gn/T//RtmpSQ4FV7/file6fc0345d6d7c'
> R(28714) malloc: *** error for object 0x7fff7a01b570: pointer being freed
> was not allocated
> *** set a breakpoint in malloc_error_break to debug
> ERROR: loading failed
> * removing ‘/Users/dsamperi/Library/R/3.0/library/Rcpp’
> * restoring previous ‘/Users/dsamperi/Library/R/3.0/library/Rcpp’
>
>
> On Thu, Oct 10, 2013 at 4:18 AM, Renaud Gaujoux <
> renaud at mancala.cbio.uct.ac.za> wrote:
>
>> Can you try with this modified version of the file?
>> I wrapped the const char* strings into std::string, maybe this will make
>> Mac happier.
>>
>> Let's communicate on this Mac issue off Rcpp-devel list for now on, since
>> it is not directly related to Rcpp, but keep rcppoctave-user (where we are
>> currently alone with Dirk :)). I will send a summary of the findings at the
>> end. People that are keen to participate can subscribe to the
>> rcppoctave-user list:
>>
>>
>> http://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcppoctave-user
>>
>>
>>
>>
>> On 10 October 2013 07:01, Dominick Samperi <djsamperi at gmail.com> wrote:
>>
>>> It appears that there is some conflict between Rcpp and Octave under
>>> Mac OS X that doesn't depend on the design of RcppOctave. Just
>>> drop the files strtest.cpp and Makevars (attached) into Rcpp/src and
>>> run R CMD INSTALL Rcpp. This leads to a memory error. The file
>>> strtest.cpp was stripped out of rcpp_octave.cpp, but it does not
>>> use RcppOctave or Rcpp (it is just a trivial Octave client).
>>>
>>> Since the problem is OS-dependent, it might have something do to
>>> with static initializers in the Octave string_vector class, or perhaps
>>> there
>>> is a MacOS-specific conflict with std::string.
>>>
>>>
>>>
>>> On Wed, Oct 9, 2013 at 4:21 AM, Renaud Gaujoux <
>>> renaud at mancala.cbio.uct.ac.za> wrote:
>>>
>>>> You are a resourceful person Dominick! :D
>>>> I was not thinking in going Mac for now, because I can't test anything,
>>>> but it will be great if we manage our way through it as well.
>>>>
>>>> 1. In src/modules/Makefile.in, the '-F' flag (for framework) is not
>>>>>      recognized by mkoctfile, so I added this temporary work-around:
>>>>>
>>>>> #R_LDFLAGS = @R_LDFLAGS@
>>>>> R_LDFLAGS = -L/Library/Frameworks/R.framework/Libraries -lR
>>>>>
>>>>
>>>> One can probably test for Mac in configure.in and tweak the R_LDFLAGS
>>>> for this platform only.
>>>>
>>>>>
>>>>> 2. Parsing the man files o_whatever.Rd fails (attempt to
>>>>>     free an unallocated pointer), so I removed all of these
>>>>>     man pages for now (the failure occurs in
>>>>>    <R>/src/library/tools/R/Rd.R, function .build_Rd_db(),
>>>>>    where .fetch_Rd_object is applied to each Rd file).
>>>>>
>>>>
>>>> The o_*.Rd files document more R-friendly wrappers to octave functions.
>>>> They contain \Sexp statements that retrieve Octave help files of the
>>>> associated function via RcppOctave o_help function:
>>>>
>>>> \Sexpr[results=rd,stage=render]{RcppOctave::o_help(<octave_function_name>,
>>>> format='rd')}
>>>>
>>>> For packages that contain such expressions in man pages, R CMD build
>>>> also generate the manual pdf, by installing the package in a temp directory
>>>> and evaluating the code in all \Sexpr statements. On Windows, R CMD INSTALL
>>>> --build also renders the man pages.
>>>> So the issue will probably solves away when the package loads properly.
>>>>
>>>> For quick testing compilation and load on Windows, I run R CMD INSTALL
>>>> directly on the package source directory with flag --no-help. To ensure a
>>>> fresh compilation at each run I use --preclean which runs the cleanup
>>>> script, and removes everything, including the .o, .so, .dll, .oct files:
>>>>
>>>> # mkdir libtest
>>>> R --vanilla CMD INSTALL -l libtest --no-help --preclean
>>>> pkg_src_directory
>>>>
>>>>
>>>>> 3. With these changes the build gets to the point where
>>>>>     the install process checks if the package can be
>>>>>     loaded, and fails with:
>>>>>
>>>>> ** testing if installed package can be loaded
>>>>> sh: line 1: 55395 Abort trap: 6
>>>>> '/Library/Frameworks/R.framework/Resources/bin/R' --no-save --slave 2>&1 <
>>>>> '/var/folders/fq/2th4vc9n1mq9cxv5nl9bmzkc0000gn/T//RtmpbjkEA7/filed5d471d0dd21'
>>>>> R(55395) malloc: *** error for object 0x7fff7684d570: pointer being
>>>>> freed was not allocated
>>>>> *** set a breakpoint in malloc_error_break to debug
>>>>> ERROR: loading failed
>>>>> * removing ‘/Users/dsamperi/Library/R/3.0/library/RcppOctave’
>>>>> * restoring previous ‘/Users/dsamperi/Library/R/3.0/library/RcppOctave’
>>>>>
>>>>> 4. If I ignore this and try to run library(RcppOctave) in R I get:
>>>>>
>>>>> > library(RcppOctave)
>>>>> Loading required package: Rcpp
>>>>> Loading required package: pkgmaker
>>>>> Loading required package: registry
>>>>> R(55403) malloc: *** error for object 0x7fff7684d570: pointer being
>>>>> freed was not allocated
>>>>> *** set a breakpoint in malloc_error_break to debug
>>>>> Abort trap: 6
>>>>>
>>>>> 5. Setting the breakpoint as instructed was not very helpful
>>>>> as the malloc code does not have symbols and cannot be
>>>>> traced.
>>>>>
>>>>>
>>>> I will double check for bad allocations in the code, but will
>>>> unfortunately be limited in my ability to debug this.
>>>>
>>>> Renaud
>>>>
>>>>
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcppoctave-user/attachments/20131010/3023608d/attachment-0001.html>