Skip to content

Configure errors say to report here... (PR#11470)

5 messages · Brian Ripley, Simon Urbanek, Adam D. I. Kramer

Original
Adam D. I. Kramer
# 
Hi,

 	I'm attempting to compile R 2.7.0 on my G5, which is running OSX
10.5.2. Yes, I know there is a precompiled binary, but my experience has
shown that if I hand-compile ATLAS on my machine and then link R against it
when compiling, R runs faster, hence me compiling it myself.

Anyway, I have this set of errors:

checking X11/Intrinsic.h usability... no
checking X11/Intrinsic.h presence... yes
configure: WARNING: X11/Intrinsic.h: present but cannot be compiled
configure: WARNING: X11/Intrinsic.h:     check for missing prerequisite headers?
configure: WARNING: X11/Intrinsic.h: see the Autoconf documentation
configure: WARNING: X11/Intrinsic.h:     section "Present But Cannot Be Compiled"
configure: WARNING: X11/Intrinsic.h: proceeding with the preprocessor's result 
configure: WARNING: X11/Intrinsic.h: in the future, the compiler will take precedence
configure: WARNING:     ## ----------------------------------- ##
configure: WARNING:     ## Report this to r-bugs at R-project.org ##
configure: WARNING:     ## ----------------------------------- ##
...
checking X11/Xmu/Atoms.h usability... no
checking X11/Xmu/Atoms.h presence... yes
configure: WARNING: X11/Xmu/Atoms.h: present but cannot be compiled
configure: WARNING: X11/Xmu/Atoms.h:     check for missing prerequisite headers?
configure: WARNING: X11/Xmu/Atoms.h: see the Autoconf documentation
configure: WARNING: X11/Xmu/Atoms.h:     section "Present But Cannot Be Compiled"
configure: WARNING: X11/Xmu/Atoms.h: proceeding with the preprocessor's result
configure: WARNING: X11/Xmu/Atoms.h: in the future, the compiler will take precedence
configure: WARNING:     ## ----------------------------------- ##
configure: WARNING:     ## Report this to r-bugs at R-project.org ##
configure: WARNING:     ## ----------------------------------- ##
...
checking tk.h usability... no
checking tk.h presence... yes
configure: WARNING: tk.h: present but cannot be compiled
configure: WARNING: tk.h:     check for missing prerequisite headers?
configure: WARNING: tk.h: see the Autoconf documentation
configure: WARNING: tk.h:     section "Present But Cannot Be Compiled"
configure: WARNING: tk.h: proceeding with the preprocessor's result
configure: WARNING: tk.h: in the future, the compiler will take precedence
configure: WARNING:     ## ----------------------------------- ##
configure: WARNING:     ## Report this to r-bugs at R-project.org ##
configure: WARNING:     ## ----------------------------------- ##

...the command line item I pass to configure is this:

feta:R-2.7.0> ./configure --prefix=/usr/local --with-x --without-R-framework --without-aqua --with-blas="/usr/local/lib/libptf77blas.a /usr/local/lib/libatlas.a" --with-lapack="/usr/local/lib/liblapack.a /usr/local/lib/libptf77blas.a /usr/local/lib/libatlas.a" CFLAGS="-maltivec -mabi=altivec -mtune=970 -faltivec -O3" CXXFLAGS="-maltivec -mabi=altivec -mtune=970 -faltivec -O3"

...the rest of configure completes fine, though when I try to build R, I
make eventually ends up failing in this manner:

gcc -std=gnu99 -I. -I../../src/include -I../../src/include -I/sw/include
-I/usr/
local/include -DHAVE_CONFIG_H   -fPIC  -maltivec -mabi=altivec -mtune=970
-falti
vec -O3 -c cpoly.c -o cpoly.o
cpoly.c: In function 'fxshft':
cpoly.c:299: error: expected identifier or '(' before 'unsigned'
cpoly.c:310: error: 'test' undeclared (first use in this function)
cpoly.c:310: error: (Each undeclared identifier is reported only once
cpoly.c:310: error: for each function it appears in.)
cpoly.c:315: error: expected expression before '__attribute__'
cpoly.c:326: error: expected expression before '__attribute__'
cpoly.c:327: error: expected expression before '__attribute__'
cpoly.c:334: error: expected expression before '__attribute__'
cpoly.c:370: error: expected expression before '__attribute__'
cpoly.c: In function 'vrshft':
cpoly.c:394: error: two or more data types in declaration specifiers
cpoly.c:394: error: expected identifier or '(' before ',' token
cpoly.c:399: error: 'b' undeclared (first use in this function)
cpoly.c:439: error: expected expression before '__attribute__'
cpoly.c:440: error: expected expression before '__attribute__'
cpoly.c:459: error: expected expression before '__attribute__'
cpoly.c:460: error: expected expression before '__attribute__'
cpoly.c:461: error: expected expression before '__attribute__'
cpoly.c:462: error: expected expression before '__attribute__'
cpoly.c: At top level:
cpoly.c:476: error: expected ';', ',' or ')' before 'unsigned'
cpoly.c:498: error: two or more data types in declaration specifiers
cpoly.c:498: error: conflicting types for 'nexth'
cpoly.c:81: error: previous declaration of 'nexth' was here
cpoly.c: In function 'nexth':
cpoly.c:498: error: parameter name omitted
cpoly.c:506: error: expected expression before '__attribute__'
make[3]: *** [cpoly.o] Error 1
make[2]: *** [R] Error 2
make[1]: *** [R] Error 1
make: *** [R] Error 1

...the one reference to a problem like this I could find via google searches
is the post at

http://osdir.com/ml/lang.r.mac/2006-10/msg00016.html

regarding R 2.4.0, which I could not find a response to.

Any help is appreciated!

Thanks,
Adam Kramer
Brian Ripley
# 
What version are your Xcode tools?  What X11 do you have installed (I see 
/sw/include in your path, and that is not normally needed)?

Although configure says to report here, it is reporting a bug in your OS, 
not in R.  There will be more information in your config.log, (and perhaps 
some of it useful).

Where did all those gcc flags come from?  I don't see them in 'R 
Installation and Administration'. Perhaps leaving them off might solve the 
problem.
On Fri, 16 May 2008, adik at ilovebacon.org wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Simon Urbanek
    

    

      
      #
    

  


  

      
Adam,

this it not a bug in R, this is merely a user error. Please don't file  
self-inflicted problems as bugs (see posting guide). For Mac-related  
questions there is R-SIG-Mac. Brian's response was to the point - you  
break the compilation with the custom flags you use. See details below.

        
On May 16, 2008, at 12:40 AM, adik at ilovebacon.org wrote:

        

            

      
      
      
    

        
FWIW that is not necessary in general, you can plug in any BLAS by  
pointing the libRblas.dylib symlink at it (or an umbrella stub created  
for it if it's incomplete).

            

      
      
      
    

        
This is likely a mess up of your headers and libraries. It's not  
causing the error, but still you're heading for trouble.

            

      
      
      
    

        
Depending on your compiler this may not work, because you're forcing  
altivec constructs which introduces new keywords to the C language. R  
doesn't use any altivec commands so it will break the compilation (as  
you can see below). [This is based on my memory - I cannot reproduce  
the problem on my G5, but then I'm using more recent tools and OS, so  
your milleage may vary].

BTW: I have no idea what you're trying to achieve with those flags,  
they are essentially no-ops, because R doesn't contain any altivec  
code (and reals are "double" in R and thus AltiVec is of no use  
anyway). If you want to optimize for speed, have a look at posts by  
Jan de Leeuw and the old OptimizingR wiki page (also see -fast in gcc  
and what it implies, although you need dynamic).

            

      
      
      
    

        
You're including /sw (Fink) - this is asking for trouble, because Fink  
contains incompatible versions of system libraries and headers (likely  
this causes the cpp problems above). Again, this is not the primary  
reason for the failure, but I hope you won't ask for help with that  
binary ;).

Cheers,
Simon

            

      
      
      
    

        
You found the right thread with the right hint, except you interpreted  
it the wrong way - altivec was needed by the Apple's Carbon headers  
(which are no longer used), but in R itself it caused the problem you  
see in the post (and thus yours).

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Adam D. I. Kramer
    

    

      
      #
    

  


  

      
Dear Dr. Ripley,

 	Thanks for your response. I'm running Xcode Tools 3.0, and the
latest Xquartz X11 (Xquartz 2.2.1, xorg-server 1.3.0-apple20) as has been
maintained by Apple's X11 mailing list (see http://xquartz.macosforge.org/ )

 	...the CFLAGS included were the same ones I was passing gfortran for
the lapack-3.1.1 and ATLAS builds, though at your suggestion I left them off
in favor of those listed at
https://stat.ethz.ch/pipermail/r-sig-mac/2005-February/001641.html for use
on the G5. I moved /sw last in my $PATH to ensure that I'm actually using
Apple's gcc, properly included /usr/local/lib/libptcblas.a, and used
--disable-R-framework instead of --without (oops). After fixing these
foolish errors, R compiles and installs correctly, and runs! Only 5-10%
faster, but hey.

 	Many thanks. Of course, if you are interested in debugging the error
based on the earlier compile (i.e., the one that told me to email this
list), I am happy to help.

--Adam

        
On Fri, 16 May 2008, Prof Brian Ripley wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Adam D. I. Kramer
    

    

      
      #
    

  


  

      
Simon,

I apologize for wasting your time. When a configure script sets three lines
of text apart to say, "Report this to r-bugs at R-project.org" on three
different occasions, I figured it was polite to do so...but I do thank you
for your help, which with Dr. Ripley's was sufficient to get me compiled and
speedy. Thanks again!

--Adam

        
On Fri, 16 May 2008, Simon Urbanek wrote: