Matthias Klose <[EMAIL PROTECTED]> wrote:

> Package: icu
> Usertag: g++-3.4
>
> Please (re)check, if the package can be built by g++, and if not, if it
> can be built using gcc-snapshot (or g++-4.1, currently found in
> experimental). Please keep the bug report open until you can avoid the
> build dependency on g++-3.4. Collect further information in this bug
> report (i.e. if it builds with a newer/experimental version).

As of today, on i386, icu 3.4 passes its test suite with g++ 3.4 and
fails with both the current gcc 4.0 in sid (4.0.2-7) and with
gcc-snapshot (20051124-1) though the nature of the test suite failures
are slightly different.  In both cases, test cases involving NaN
fail, though more pass with 4.1 than 4.0.

After doing a little bit of digging, I found out that using
-fno-strict-aliasing with gcc 4.0 works around the gcc 4.0 bug that
breaks icu 3.4, and based on stuff I see going on upstream, it appears
that the underlying issue is being addressed, so this workaround
should be temporary.  (In other words, when icu 3.6 comes out, it
should no longer be necessary.)  I have manually run the test suite on
ia64, powerpc, and i386 (thus covering a both byte orders and both
word sizes), and it passes on those platforms.  I'm going to go ahead
and upload a new icu that builds with just plain gcc and g++ (so
4.0.2) using -fno-strict-aliasing today.

-- 
Jay Berkenbilt <[EMAIL PROTECTED]>


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to