[Rd] Does R use "computed gotos" - a gcc extension of C?

[luke-tierney at uiowa.edu]

On Sat, 5 Mar 2011, Dr. David Kirkby wrote:

> On 03/ 4/11 11:40 PM, luke-tierney at uiowa.edu wrote:
>> On Fri, 4 Mar 2011, Dr. David Kirkby wrote:
>> 
>>> The R manual says R will not build with gcc on 64-bit Solaris x86 with
>>> gcc
>>> 
>>> http://cran.r-project.org/doc/manuals/R-admin.html#Solaris
>>> 
>>> "Tests with gcc32 on ‘x86’ and ‘amd64’ have been less successful:
>>> ‘x86’ builds have failed on tests using complex arithmetic33, whereas
>>> on ‘amd64’ the builds have failed to complete in several different
>>> ways, most recently with relocation errors for libRblas.so. "
>>> 
>>> I know what the "relocation errors" problem is. That library (and in
>>> fact two other R libraries) all have non-PIC code in them, despite the
>>> fact the source is compiled with the -fPIC option.
>>> 
>>> http://blogs.sun.com/rie/entry/my_relocations_don_t_fit
>>> 
>>> shows how to prove this. If one runs this command on Solaris:
>>> 
>>> $ elfdump -d libRblas.so | fgrep TEXTREL
>>> 
>>> there is some output showing that theres non-PIC code present in the R
>>> library.
>>> 
>>> R is compiled with -fPIC on Solaris, but certain things can cause
>>> non-PIC code to be generated even with that option. One is by the use
>>> of "computed gotos" which is a gcc extension. I'm wondering if R uses
>>> any of these.
>> 
>> Yes -- in the byte code interpreter in eval.c
>> 
>> luke
>
> Thank you Luke. Do you know if there may be any others?

I do not.

> Do you know if that bit of code gets compiled into all 3 of the R libraries? 
> I tried replacing

I do not know

> #define NEXT() (__extension__ ({goto *(*pc++).v;}))
>
>
> by a function which did absolutely nothing. The code built, but still had the 
> library issues.
>
> I'm almost certain that the definition of NEXT will cause problems, but I'm 
> not convinced it is the only issue.
>
> What happens if a non-GNU compiler is used? I assume these GNU extensions 
> don't get used, so how comes the code builds? Is there any way I can disable 
> the use of the GNU extensions, while still building with gcc.

This particular bit It only uses the gcc extensions for gcc compilers,
and this can be turned off by defining NO_THREADED_CODE.

>
> It is rather annoying that the code has __extension__ in it, which disables 
> the warnings about the use of GCC extensions.
>
> http://gcc.gnu.org/onlinedocs/gcc/Alternate-Keywords.html
>
> Why is there a need to hide the use of the extensions?

To avoid spurious warnings

> I'd personally like to 
> see just standard C used, without any extensions. Then problems like I'm 
> having would be less likely to occur.

This extention is very useful for implementing byte code interpreters,
which is why it is being used.  As mentioned above, its use can be
turned off.

luke

-- 
Luke Tierney
Statistics and Actuarial Science
Ralph E. Wareham Professor of Mathematical Sciences
University of Iowa                  Phone:             319-335-3386
Department of Statistics and        Fax:               319-335-3017
    Actuarial Science
241 Schaeffer Hall                  email:      luke at stat.uiowa.edu
Iowa City, IA 52242                 WWW:  http://www.stat.uiowa.edu