Re: [Rd] Package crashes R under Windows Vista
Dear Duncan,
I know that I am asking an unusual question, since MS VC++ is not
supported, and I have of course installed the Rtools. However, asking
support from MS is not an option for the following reason:
Looking at different VC forums everybody tells you that you should
compile all libraries with the same option and you should not mix
options /MD with /MT, and this is the answer I will get.
Furthermore, as I have already mentioned, library xps.dll compiled with
/MD works perfectly fine on WinXP and Vista when used independent from
R. The problem is that when calling library(xps) from within R, R can
load library xps.dll, but it cannot load the libraries which are called
from library xps.dll such as library libCore.dll.
Since the question is why can R not load a C++ library which is called
from library xps.dll, asking this question at R-devel seems for me to be
the only option.
Best regards
Christian
Duncan Murdoch wrote:
On 18/09/2009 6:12 PM, cstrato wrote:
Dear all,
Please, please I really need your help, since I am not able to solve
the problem myself:
It seems that until now most Windows users of my Bioconductor package
"xps" use Windows XP, but recently a user tried to use it on Windows
Vista where it crashed R immediately. Since I assume that some users
may migrate to the upcoming Windows 7 where the problem will be the
same, I would appreciate any help to be able to support Vista/Windows
7, thank you in advance.
You are requiring specific Microsoft tools to compile a package, which
fails on one Microsoft OS, but not another. Isn't this a question you
should address to Microsoft?
For the record:
We support compiling in the compiler listed in the R-admin manual,
which is available from www.murdoch-sutherland.com/Rtools. If you
choose to use a different compiler, you should seek support from its
supplier.
Duncan Murdoch
The problem ist that my package consists mainly of C++ code which I
must compile with Visual C++ (Express Edition 2008), since it depends
on the ROOT C++ framework. Windows users download and install first
the ROOT Windows binary compiled with VC9 and then install the
Windows binary of xps.
Thus package xps contains a "Makefile.win" where I use one of the
compiler options "/MT" or "/MD", respectively, see:
http://msdn.microsoft.com/en-us/library/2kzt1wy3.aspx
Since the Windows binary of ROOT is compiled with option "/MD" I
should in principle use this option, however, R does not accept this
option. Luckily, for Winodws XP I can compile xps with "/MT" and this
works perfectly fine.
The real problem is that neither option works on Windows Vista,
although I can run "R CMD INSTALL" in both cases. However, when
loading "library(xps)" using Rterm.exe I get the following two cases:
1, xps compiled on Vista with option "/MD":
new R session:
> library(xps,lib.loc="C:/home/Rabbitus/CRAN/myRlib")
Error in inDL(x, as.logical(local), as.logical(now), ...) :
cannot load shared library
'C:/home/CS/CRAN/myRlib/xps/libs/xps.dll':
LoadLibrary failure: A DLL-Initialisationroutine failed.
Then R crashes and Vista displays a Dialogbox with the following
message:
R for Windows terminal front-end does no longer function
Anwendungszeitstempel:4a928a5c
Fehlermodulname:libCore.dll_unloaded
Fehlermodulversion:0.0.0.0
Fehlermodulzeitstempel:4a4b94a1
Ausnahmecode:c005
Ausnahmeoffset:0413a6f0
Betriebsystemversion:6.0.6001.2.1.0.768.3
Gebietsschema-ID:1031
Zusatzinformation 1:b2be
Zusatzinformation 2:82109204b3973566ace818217390e302
Zusatzinformation 3:1563
Zusatzinformation 4:deee40b8797d25628c21b9ef47e8a50c
2, xps compiled on Vista with option "/MT":
new R session:
> library(xps,lib.loc="C:/home/CS/CRAN/myRlib")
> scheme.test3 <-
root.scheme(paste(.path.package("xps"),"schemes/SchemeTest3.root",sep="/"))
> data.test3 <- import.data(scheme.test3, "DataTest3",
celdir=paste(.path.package("xps"),"raw",sep="/"))
==
=== STACKTRACE ===
==
==
= END STACKTRACE =
==
Then R crashes and Vista displays a Dialogbox with the following
message:
R for Windows terminal front-end does no longer function
Problemsignatur:
Problemereignisname:APPCRASH
Anwendungsname:Rterm.exe
Anwendungsversion:2.92.49384.0
Anwendungszeitstempel:4a928a5c
Fehlermodulname:ntdll.dll
Fehlermodulversion:6.0.6001.18000
Fehlermodulzeitstempel:4791a7a6
Ausnahmecode:c005
Ausnahmeoffset:0006814c
Betriebsystemversion:6.0.6001.2.1.0.768.3
Gebietsschema-ID:1031
Zusatzinformation 1:92ec
Zusatzinformation 2:c556221fb8ba46fdaae5867fc20d6575
Zusatzinformation 3:c65f
Zus
[Rd] Return a list from a .Call but segfaults
Hello,
I call a function via .Call passing to it a raw vector(D) and an
integer(I)
The vector is a series K1,KData1, V1,VData1, K2, KData2, ...
where the integer K1 is the length of Data1 and similarly for Ki (wrt
Datai)(similarly for V*) There 2*I such pairs( (Ki,KDatai), (Vi,VDatai))
The numbers Ki(and Vi) are written in network order.
I am returning a list of I elements each element a list of two
elements corresponding to KData and VData (which are serialized R
objects) .
ll <- .Call("returnListOfKV", rawkv, numread)
When use the function with browser(), and type head(ll) i get valid
results.
However, i get a segfault when the code runs the next line
j <- append(j,ll)
(where j was defined as vector(mode='list') )
The code below looks alright, but it appears to crash when getting
copied.
Now one could blame message2rexp (in kk_), but once in R, the result
of .Call is assigned to ll, everything should be protected.
In fact, if in kk_ i simply return a VECSXP with 2 elements, no
errors, so one could conclude that message2rexp is to blame.
Yet, when i replace the code with R code, that is using readBin and
only calling calling C(to call message2rexp) to deserialize a raw
vector, I get no such error.
There is something I'm missing , could someone tell me where?
Thank you for your time
Regards
Saptarshi
=== Using GDB?
I would like to run it through gdb, i get the following message
[Thread debugging using libthread_db enabled]
Error while reading shared library symbols:
Cannot find new threads: generic error
Cannot find new threads: generic error]
and cannot proceed.
Code
SEXP returnListOfKV(SEXP raw,SEXP numread){
if(TYPEOF(raw)!=RAWSXP){
return(R_NilValue);
}
int num = INTEGER(numread)[0];
char *rawdata = (char*)RAW(raw);
SEXP KV ,rval;
int r;
PROTECT(rval = Rf_allocVector(VECSXP, num));
for(int i=0;i r = reverseUInt(*((int*) rawdata)); //converts network order to
host
rawdata+=4; //yes, hard code size of int
PROTECT(k= kk_(rawdata,r)); //deserializes data and returns a
SEXP
rawdata+= r;
SET_VECTOR_ELT(KV,0, k);
UNPROTECT(1);
r = reverseUInt(*((int*) rawdata));
rawdata+=4;
PROTECT(k= kk_(rawdata,r));
rawdata+=r;
SET_VECTOR_ELT(KV,1, k);
UNPROTECT(1);
SET_VECTOR_ELT(rval,i,KV);
UNPROTECT(1);
}
UNPROTECT(1);
return(rval);
}
SEXP kk_(char *d,int n){
SEXP k;
REXP *rexp = new REXP();
rexp->Clear();
rexp->ParseFromArray(d,n);
PROTECT(k = message2rexp(*rexp));
delete(rexp);
UNPROTECT(1);
return(k);
}
Saptarshi Guha | saptarshi.g...@gmail.com | http://www.stat.purdue.edu/~sguha
__
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
Re: [Rd] Return a list from a .Call but segfaults
After some examination, this is what i found (i changed the code
somewhat, i dont modify rawdata anymore)
load("/tmp/v.Rdata")
dyn.load("/ln/meraki/custom/lib64/R/library/mylib/libs/lib.so")
v=.Call("returnListOfKV",rawkv,numread)
## v is a list of 28231 lists each of length 2
j=v[1:1] ; u=list();u=append(u,j)
and
j=v[1:15164] ; u=list();u=append(u,j)
both work but for n> 15164
j=v[1:n] ; u=list();u=append(u,j)
returns
Error: protect(): protection stack overflow
I have no idea why...
Regards
Saptarshi
On Sat, Sep 19, 2009 at 10:15 PM, Saptarshi Guha
wrote:
> Hello,
> I call a function via .Call passing to it a raw vector(D) and an integer(I)
> The vector is a series K1,KData1, V1,VData1, K2, KData2, ...
> where the integer K1 is the length of Data1 and similarly for Ki (wrt
> Datai)(similarly for V*) There 2*I such pairs( (Ki,KDatai), (Vi,VDatai))
>
> The numbers Ki(and Vi) are written in network order.
>
> I am returning a list of I elements each element a list of two elements
> corresponding to KData and VData (which are serialized R objects) .
>
> ll <- .Call("returnListOfKV", rawkv, numread)
>
> When use the function with browser(), and type head(ll) i get valid results.
> However, i get a segfault when the code runs the next line
>
> j <- append(j,ll)
>
> (where j was defined as vector(mode='list') )
>
> The code below looks alright, but it appears to crash when getting copied.
> Now one could blame message2rexp (in kk_), but once in R, the result of
> .Call is assigned to ll, everything should be protected.
> In fact, if in kk_ i simply return a VECSXP with 2 elements, no errors, so
> one could conclude that message2rexp is to blame.
>
> Yet, when i replace the code with R code, that is using readBin and only
> calling calling C(to call message2rexp) to deserialize a raw vector, I get
> no such error.
>
> There is something I'm missing , could someone tell me where?
> Thank you for your time
> Regards
> Saptarshi
>
>
> === Using GDB?
> I would like to run it through gdb, i get the following message
> [Thread debugging using libthread_db enabled]
> Error while reading shared library symbols:
> Cannot find new threads: generic error
> Cannot find new threads: generic error]
>
> and cannot proceed.
>
> Code
> SEXP returnListOfKV(SEXP raw,SEXP numread){
>
> if(TYPEOF(raw)!=RAWSXP){
> return(R_NilValue);
> }
> int num = INTEGER(numread)[0];
> char *rawdata = (char*)RAW(raw);
> SEXP KV ,rval;
> int r;
> PROTECT(rval = Rf_allocVector(VECSXP, num));
> for(int i=0;i SEXP k = R_NilValue;
> PROTECT(KV = Rf_allocVector(VECSXP, 2));
>
> r = reverseUInt(*((int*) rawdata)); //converts network order to host
> rawdata+=4; //yes, hard code size of int
> PROTECT(k= kk_(rawdata,r)); //deserializes data and returns a SEXP
> rawdata+= r;
> SET_VECTOR_ELT(KV,0, k);
> UNPROTECT(1);
>
> r = reverseUInt(*((int*) rawdata));
> rawdata+=4;
> PROTECT(k= kk_(rawdata,r));
> rawdata+=r;
> SET_VECTOR_ELT(KV,1, k);
> UNPROTECT(1);
>
> SET_VECTOR_ELT(rval,i,KV);
> UNPROTECT(1);
> }
> UNPROTECT(1);
> return(rval);
> }
>
>
> SEXP kk_(char *d,int n){
> SEXP k;
> REXP *rexp = new REXP();
> rexp->Clear();
> rexp->ParseFromArray(d,n);
> PROTECT(k = message2rexp(*rexp));
> delete(rexp);
> UNPROTECT(1);
> return(k);
> }
>
> Saptarshi Guha | saptarshi.g...@gmail.com |
> http://www.stat.purdue.edu/~sguha
>
>
>
__
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
