Rcpp 0.11.5 test failures with new Rtools toolchain

[kevinushey]

Posting here for posterity; will try to investigate further ASAP...

Executing test function test.attributes  ... make: *** No rule to make target `C:\Users\Kevin\R-src\trunk\library\Rcpp\unitTests\cpp\attributes.o', needed by `s
ourceCpp_44614.dll'.  Stop.
Warning message:
running command 'make -f "C:/Users/Kevin/R-src/trunk/etc/x64/Makeconf" -f "C:/Users/Kevin/R-src/trunk/share/make/winshlib.mk" SHLIB_LDFLAGS='$(SHLIB_CXXLDFLAGS)
' SHLIB_LD='$(SHLIB_CXXLD)' SHLIB="sourceCpp_44614.dll" WIN=64 TCLBIN=64 OBJECTS="C:\Users\Kevin\R-src\trunk\library\Rcpp\unitTests\cpp\attributes.o attributes.
o"' had status 2

Error in sourceCpp(file.path(pathToRcppTests, "cpp", file)) :
  Error 1 occurred building shared library.

After this failure, R.exe spins here:

Executing test function test.DataFrame.AttributeProxy  ...

[kevinushey]

The first issue here is that sourceCpp("cpp/attributes.cpp") fails, for the unit tests, on Windows (for me). The following call to R CMD SHLIB is generated for me:

C:/Users/Kevin/R-src/trunk/bin/x64/R CMD SHLIB -o "sourceCpp_58407.dll" "C:\Users\Kevin\git\Rcpp\inst\unitTests\cpp\attributes.cpp" "attributes.cpp"

@jjallaire, it looks like this is related to the sourceCpp() 'find -recursive-dependencies' behaviour added in -- does that sound correct? The double-compilation of attributes.cpp seems dubious.

In particular, on OS X, I see only a single include of attributes.cpp in that statement:

/Library/Frameworks/R.framework/Resources/bin/R CMD SHLIB -o 'sourceCpp_30189.so' --preclean  'attributes.cpp' 

And, using Rcpp 0.11.4, and running sourceCpp on the old attributes.cpp file, seems to work fine.

[jjallaire]

Yes that sounds right to me. There is someplace where I try to de-dup
translation units but it looks like this isn't kicking in for the main
translation unit

On Mar 15, 2015, at 3:45 PM, Kevin Ushey notifications@github.com wrote:

The first issue here is that sourceCpp("cpp/attributes.cpp") fails, for the
unit tests, on Windows (for me). The following call to R CMD SHLIB is
generated for me:

C:/Users/Kevin/R-src/trunk/bin/x64/R CMD SHLIB -o
"sourceCpp_58407.dll"
"C:\Users\Kevin\git\Rcpp\inst\unitTests\cpp\attributes.cpp"
"attributes.cpp"

@jjallaire https://github.com/jjallaire, it looks like this is related to
the sourceCpp() 'find -recursive-dependencies' behaviour added in -- does
that sound correct?

In particular, using Rcpp 0.11.4, and running sourceCpp on the old
attributes.cpp file, seems to work fine.

—
Reply to this email directly or view it on GitHub
#277 (comment).

[jjallaire]

I've submitted a fix so that the test now passes (it was passing before on
other platforms but not Windows).

On Sun, Mar 15, 2015 at 4:04 PM, JJ Allaire jj@rstudio.com wrote:

Yes that sounds right to me. There is someplace where I try to de-dup
translation units but it looks like this isn't kicking in for the main
translation unit

On Mar 15, 2015, at 3:45 PM, Kevin Ushey notifications@github.com wrote:

The first issue here is that sourceCpp("cpp/attributes.cpp") fails, for
the unit tests, on Windows (for me). The following call to R CMD SHLIB is
generated for me:

C:/Users/Kevin/R-src/trunk/bin/x64/R CMD SHLIB -o "sourceCpp_58407.dll" "C:\Users\Kevin\git\Rcpp\inst\unitTests\cpp\attributes.cpp" "attributes.cpp"

@jjallaire https://github.com/jjallaire, it looks like this is related
to the sourceCpp() 'find -recursive-dependencies' behaviour added in --
does that sound correct?

In particular, using Rcpp 0.11.4, and running sourceCpp on the old
attributes.cpp file, seems to work fine.

—
Reply to this email directly or view it on GitHub
#277 (comment).

[coatless]

This issue was addressed in PR #274.

Please close. (Tagged #506)