[Rd] Rcmd SHLIB error

[Duncan Murdoch]

On 19/12/2010 10:04 AM, Gabor Grothendieck wrote:
> On Sun, Dec 19, 2010 at 9:07 AM, Duncan Murdoch
> <murdoch.duncan at gmail.com>  wrote:
>> On 19/12/2010 7:01 AM, Gabor Grothendieck wrote:
>>>
>>> On Sun, Dec 19, 2010 at 6:39 AM, Duncan Murdoch
>>> <murdoch.duncan at gmail.com>    wrote:
>>>>
>>>> On 19/12/2010 6:06 AM, Gabor Grothendieck wrote:
>>>>>
>>>>> On Sat, Dec 18, 2010 at 8:23 PM, Dominick Samperi<djsamperi at gmail.com>
>>>>>   wrote:
>>>>>>
>>>>>>
>>>>>> On Sat, Dec 18, 2010 at 5:24 PM, Gabor Grothendieck
>>>>>> <ggrothendieck at gmail.com>      wrote:
>>>>>>>
>>>>>>> I am getting this error message when I try to run Rcmd SHLIB myprog.c.
>>>>>>>   There appears to be a missing / between etc and i386 in the path.  I
>>>>>>> am on Windows Vista and am using R version 2.12.1 Patched (2010-12-16
>>>>>>> r53864) and just downloaded Rtools 2.12 today.   Is this a bug in R?
>>>>>>> How can I resolve this?
>>>>>>
>>>>>> This is due to new behavior in CYGWIN (the basis for Rtools). The
>>>>>> new CYGWIN does not like non-standard paths that have C: in them.
>>>>>> Some CYGWIN programs will not work when fed file names using
>>>>>> this syntax, but normally this is just a warning. The warning can be
>>>>>> suppressed by using the Control Panel to set CYGWIN=nodosfilewarning.
>>>>>> Another solution is to use POSIX equivalent paths of the form
>>>>>> /cydrive/c/PROG..., but this would require more extensive changes
>>>>>> to build scripts, etc.
>>>>>>
>>>>>> There is another change that causes Rtools tar to choke in some
>>>>>> circumstances under Windows because it tries to change ownership
>>>>>> of the files extracted when it shouldn't. These error
>>>>>> messags can be suppressed by using the Control Panel to set
>>>>>> TAR_OPTIONS=--no-same-owner.
>>>>>>
>>>>>
>>>>> I am still wondering what to do about this?  I have created a
>>>>> directory called C:\PROGRA~1\R\R-212~1.X\etci386 and have copied
>>>>> Makeconf into it.  That lets me use Rcmd SHLIB but its quite an ugly
>>>>> workaround and there must be some better resolution than that.  Also I
>>>>> am wondering if this same problem will crop up elsewhere.
>>>>>
>>>>
>>>> You could try running within Cygwin, as I do, on the theory that what
>>>> works
>>>> for me might work for you.  I do use the "nodosfilewarning" setting as
>>>> Dominick mentioned.
>>>>
>>>> However, I will say that my Windows build system is much less reliable
>>>> than
>>>> it used to be.  I don't know the cause:  I started 64 bit builds on
>>>> Windows
>>>> 7 at around the same time as I upgraded Cygwin, so it could be any of
>>>> those
>>>> causes.  I'm tending to believe it's a Cygwin problem or a Cygwin-Win7
>>>> incompatibility, but I haven't had time to work out a reliable system.
>>>>   (Symptoms I see are that very few overnight builds are succeeding;
>>>> almost
>>>> all the ones you see on CRAN/bin/windows/base have required manual
>>>> restarts
>>>> after they've failed partway through.)
>>>>
>>>> Unfortunately, I am not going to have the time to diagnose or fix this in
>>>> the next couple of months.  If you have spare time, you might want to try
>>>> older versions of the Rtools (perhaps mixing new compilers with old
>>>> Rtools/bin and Cygwin DLLs).
>>>>
>>>> Duncan Murdoch
>>>>
>>>
>>> Where in the sources is the C:\PROGRA~1\R\R-212~1.X\etci386 [sic] path
>>> set?
>>
>> I don't know.  Possibilities include src/gnuwin32/fixed/Makefile line 31
>> (where a bad R_ARCH would do it).  Do you have R_ARCH set as an environment
>> variable or on the command line when you call make? I think the latter would
>> override the setting in MkRules, I'm not sure about the former.
>>
>
> Thanks.  I now have a better workaround.  The batch file shown below
> works but if /i386 is changed to i386 then it fails because it looks
> for a directory called etci386:
>
>    setlocal
>    set R_ARCH=/i386
>    "C:\Program Files\R\R-2.12.x\bin\i386\Rcmd.exe" SHLIB myprog.c
>    endlocal
>
> 1. It seems the / is missing between etc and $(R_ARCH) on lines 31 and
> 38 of that Makefile (or if etci386 is intended then the actual
> structure of R that is installed is wrong).

No, the slash is supposed to be part of R_ARCH, which isn't a variable 
you should need to set.  It is set in MkRules, as I mentioned above.

> 2. page 8 of R-admin manual says that there is an --arch argument with
> values i386, 32, x64, 64 but neither "Rcmd --help" nor "Rcmd SHLIB
> --help" document an argument for setting the architecture.

That argument appears to work for me.  If you want to submit a patch so 
--help displays it, I'll take a look.

> 3. on page 12 of the R-ext manual it refers to
> R_HOME/etcR_ARCH/Makeconf so the documentation and how the Makefile
> works seems inconsistent with the actual R directory structure which
> is etc/i386 and not etci386.
>

See 1.

Duncan Murdoch