[R-pkg-devel] debugging problems with valgrind
i am developing a little package that basically is a simple wrapper around some C++ code, that is being called via the RCpp package. Basically the code works, but i did find out that the results are different for 32bit and 64bit environments, so this might point to problems in the C++ code. The 32bit box is a virtual machine, the 64bit is running native. Both systems are Ubuntu 15.04 with R 3.1.2. Also, when calling build_win() i get different results than on my computer. To debug the situation, i tried to use valgrind as a debugger on the 32bit virtual machine. So instead of a simple "R --vanilla <./demo.R" (the demo calls only the corresponding unit test) i called now "R -d "valgrind" --vanilla < demo.R". The (very funny) result is that when called with valgrind, my tests are now running fine, i.e. i get the very same results as on my 64bit box. So in some ways valgrind seems to change the behaviour of R, luckily into a 'good' direction-- but which is nothing that i would have expected. (To make sure, i also did the same two commands on my native 64bit system, there is no such change, i.e. valgrind does not change anything) I am not sure if i can make a minimal test script that reproduces the behaviour. Before i try i would like to hear if anyone ever encountered something similar with using valgrind and R, and maybe can point out if this is something related to R, valgrind, Rcpp, C++ code or something completely different (while i am certain that the C++ code is not really 64bit ready yet). (In case anyone wants to peek into the code, it can be found at https://github.com/aydindemircioglu/yakmoR) Thanks, Aydin [[alternative HTML version deleted]] __ R-package-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-package-devel
[R-pkg-devel] how to reproduce an ASAN container overflow
dear all,
i am trying for three days to reproduce the container overflow in my
package,
alas i am not able to. for the full output see:
http://www.stats.ox.ac.uk/pub/bdr/memtests/clang-ASAN/lasvmR/00check.log
the error seems actually to be easy (vector::clear before std::copy), but i
want
to be sure that the cran-check does not fail again. so i'd like to
reproduce the error
first. from dirks tutorial (
http://dirk.eddelbuettel.com/code/sanitizers.html),
i would assume that if the sanitiziers package work, i'd be good,
i.e. running docker yields the following on my ubuntu 15.10 (alpha):
~$ docker run --rm -ti -v $(pwd):/mnt rocker/r-devel-san /bin/bash
root@126d39bbe93f:/# Rdevel -e
"install.packages('sanitizers');library(sanitizers);
print(stackAddressSanitize(42))"
--- CUT ---
stack_address.cpp:16:32: runtime error: index 142 out of bounds for type
'int [100]'
stack_address.cpp:16:33: runtime error: load of address 0x7ffde3477988 with
insufficient space for an object of type 'int'
0x7ffde3477988: note: pointer points here
d0 61 00 00 e0 7c b4 02 d0 61 00 00 b3 8a b5 41 00 00 00 00 60 5e f2 ff
22 7f 00 00 c0 61 99 ff
^
[1] 45382880
>
=
==11==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 6 byte(s) in 2 object(s) allocated from:
#0 0x7f2300e8537a in malloc
(/usr/lib/x86_64-linux-gnu/libasan.so.2+0x9437a)
#1 0x7f22fd8ba9e9 in strdup (/lib/x86_64-linux-gnu/libc.so.6+0x819e9)
SUMMARY: AddressSanitizer: 6 byte(s) leaked in 2 allocation(s).
root@126d39bbe93f:/#
so checking my program in this docker environment with Rdevel should more
or less reproduce the cran report above. but it does not. so "Rdevel CMD
build ."
and then "Rdevel CMD check --as-cran ./lasvmR_0.1.1.tar.gz" yields no
relevant
error:
$ Rdevel CMD check --as-cran ./lasvmR_0.1.1.tar.gz
--- CUT ---
* checking for unstated dependencies in examples ... WARNING
summary.c:876:5: runtime error: null pointer passed as argument 2, which is
declared to never be null
* checking line endings in C/C++/Fortran sources/headers ... OK
* checking compiled code ... OK
WARNING
‘qpdf’ is needed for checks on size reduction of PDFs
* checking installed files from ‘inst/doc’ ... OK
* checking files in ‘vignettes’ ... OK
* checking examples ... OK
--- CUT ---
=
==602==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 72128 byte(s) in 2 object(s) allocated from:
#0 0x7f3836fca4e1 in __interceptor_calloc
(/usr/lib/x86_64-linux-gnu/libasan.so.2+0x944e1)
#1 0x7f382a3e08d9 (/usr/lib/x86_64-linux-gnu/libcurl.so.4+0x1e8d9)
#2 0xf98996c57a955ff ()
--- CUT ---
the compile flags in 00install.out seems also ok for me (with no error or
warnings)
--- CUT ---
g++ -fsanitize=address,undefined -I/usr/local/lib/R/include -DNDEBUG
-I/usr/local/include -I"/usr/local/lib/R/site-library/Rcpp/include"
-fpic -pipe -Wall -pedantic -O3 -c vector.cpp -o vector.o
g++ -fsanitize=address,undefined -shared -L/usr/local/lib/R/lib
-L/usr/local/lib -o lasvmR.so RcppExports.o kcache.o kernel.o la_svm.o
la_test.o lasvm.o lasvm_wrapper.o messages.o vector.o
-L/usr/local/lib/R/lib -lR
--- CUT ---
i also tried to recompile R-devel from svn locally on my machine with
various
-fsanitize options and exchanging gcc with clang (this i could not do on
the
docker image, due to some apt-get version errors), taking the configure
switches
from dirks docker container. but whatever i do, finally i only get either
no errors
at all or i get a bunch of errors from other packages that makes R CMD stop
before getting to compiling/checking my package.
can someone tell me what i do wrong and how to get the exact same error
report
as the cran check above? i also gladly accept RTFM statements, i could not
find
anything relevant and to the point, or a page where the exact configuration
and/or compile flags for the cran-report can be found.
sorry if i missed any important information you might need to help me.
all the best,
aydin
[[alternative HTML version deleted]]
__
R-package-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel
[R-pkg-devel] compiling and executing binaries while testing
hi, i am sorry if this has been asked already, but i could not find an answer. in short: am i allowed to download, compile and execute (github) packages while testing? reason for this: the package i develop is a simple 'bridge', the goal is to allow users to easily access (commonly used or their own) machine learning binaries (like LIBSVM). obviously i will not add any kind of binary to the package itself, but on the other hand i cannot test the functionality of the package without having a binary. therefore it seems most reasonable to compile the binary while testing and delete it afterwards. are there any objections to this procedure? thanks & all the best, aydin __ R-package-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-package-devel
