regression: g++ produces spurious alignment errors for prototypes

[Jens Gustedt]

Binary package hint: gcc

This appeared recently when upgrading from gutsy to intrepid. Compiling the following with g++

void toto1(double*const __attribute__((aligned(16))) X);
void toto2(double* const (&X)[1]);

produdes

g++ -c -o test.o test.cc
test.cc:2: error: alignment of array elements is greater than element size

Observe the `2' in the error message, the error is reported for toto2.
(For toto1 an error would be acceptable, I think).

But even worse, if we take

void toto0(double* const (&X)[1]);
void toto1(double*const __attribute__((aligned(16))) X);
void toto2(double* const (&X)[1]);

All three prototypes compile flawlessly.

This particular testing was done with 4:4.3.1-1ubuntu2 on an 8.10 amd64,
but we observed the same thing on 8.10 i386
Previously, everything was working well on 8.04.

ProblemType: Bug
Architecture: amd64
DistroRelease: Ubuntu 8.10
Package: g++ 4:4.3.1-1ubuntu2
ProcEnviron:
 SHELL=/bin/tcsh
 PATH=/home/username/bin:/usr/bin:/usr/sbin:/bin:/usr/X11R6/bin:/usr/bin/X11:/usr/local/bin:/usr/local/bin
 LANG=de_DE.iso885915
SourcePackage: gcc-defaults
Uname: Linux 2.6.27.5 x86_64

[Jens Gustedt (jens-gustedt)]

[This is a copy of a bug that I posted a year ago on ubuntu's launchpad.
Nothing happened there, and now I am still observing the same problem
with gcc 4.4.1 on a newly upgraded ubuntu 9.10]

Compiling the following with g++

void toto1(double*const __attribute__((aligned(16))) X);
void toto2(double* const (&X)[1]);

produdes

g++ -c -o test.o test.cc
test.cc:2: error: alignment of array elements is greater than element size

Observe the `2' in the error message, the error is reported for toto2.
(For toto1 an error would be acceptable, I think).

But even worse, if we take

void toto0(double* const (&X)[1]);
void toto1(double*const __attribute__((aligned(16))) X);
void toto2(double* const (&X)[1]);

All three prototypes compile flawlessly.

[Jens Gustedt (jens-gustedt)]

Still present in gcc 4.4.1 of ubuntu 9.10
I know opened a bug on gcc bugzilla:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41966

Jens

[Paolo-carlini (paolo-carlini)]

No errors in 4_6-branch and mainline.

[Jens Gustedt (jens-gustedt)]

Just to add to the list, since this finally found some attention.
The problem is also manifest for 4.5.2, only the difference is that this time the three line version at the end of the original report also have the error.