Bug 445606 – sys-devel/gcc-4.5.x: Preprocessor munges the word "vector" when -maltivec is in effect

• Status: RESOLVED FIXED
• Product: Gentoo Linux

[Description Marcus Comstedt 2012-12-02 13:28:00 UTC]

When preprocessing the following snippet:

namespace a {
  typedef int vector[3];
};
typedef a::vector vector;

using g++ -maltivec -E foo.cpp, then the output is

# 1 "foo.cpp"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "foo.cpp"
namespace a {
  typedef int vector[3];
};
typedef a::vector;

Notice how the final "vector" has disappeared.  It seems that gcc is providing a magical #define for "vector" which is not working correctly.  Using -dD, I can see

#define vector vector

but if I actually add this definition (or #undef vector) to the top of the file, then the output becomes correct.

g++ (Gentoo 4.5.4 p1.0, pie-0.4.7) 4.5.4


Reproducible: Always

Steps to Reproduce:
1. Type the above example into a file
2. Preprocess it with g++ -maltivec -E
3. Observe the result
Actual Results:  
typedef a::vector;


Expected Results:  
typedef a::vector vector;

[Comment 1 Marcus Comstedt 2012-12-02 15:03:44 UTC]

Another way of getting the correct output is this:

namespace a {
  typedef int vector[3];
};
#define DUMMY
typedef a::vector DUMMY vector;

[Comment 2 SpanKY 2012-12-02 20:27:27 UTC]

not a bug.  "vector" is a gcc builtin type.

[Comment 3 Marcus Comstedt 2012-12-02 20:29:40 UTC]

So you are saying that it is correct that

namespace a {
  typedef int vector[3];
};
typedef a::vector vector;

and

namespace a {
  typedef int vector[3];
};
#define DUMMY
typedef a::vector DUMMY vector;

yield different results?  And since when is the preprocessor concerned with types?

[Comment 4 SpanKY 2012-12-02 20:40:29 UTC]

feel free to file a bug at gcc.gnu.org and cc toolchain@g.o.  i've dealt with enough people getting yelling at "vector" and "__vector" to waste time on more.

[Comment 5 Marcus Comstedt 2012-12-02 21:25:21 UTC]

Actually the problem seems to have been reported (and fixed, so much for "not a bug" :) upstreams already:  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48226

So either the fix could be backported to 4.5.4, or 4.7.2 (which has the fix) unmasked for ppc(64).

[Comment 6 Marcus Comstedt 2012-12-03 15:52:08 UTC]

For the record:  The patch from the gcc bugzilla applies cleanly to gcc-4.5.4 (I put it in my /etc/portage/patches), and the resulting gcc does not exhibit the problem.

[Comment 7 SpanKY 2012-12-03 21:24:21 UTC]

(In reply to comment #5)

we don't really do arch-specific toolchain unmasking anymore.  especially not gcc-4.7 right now.

if there's a patch available, it's easy enough to add to the 4.5.x patchset.

[Comment 8 SpanKY 2012-12-21 05:19:40 UTC]

queued up in all 4.5.x and 4.6.x patch series, and pushed out in current 4.5.4 and 4.6.3 ebuilds

[Comment 9 Marcus Comstedt 2012-12-21 10:37:12 UTC]

Thanks!

I confirm the issue as being resolved by the latest ebuild.

Happy holidays!