On 2/6/2013 3:35 AM, Richard Hipp wrote:
On Wed, Feb 6, 2013 at 3:48 AM, Edward Berner <[email protected]
<mailto:[email protected]>> wrote:
(I should mention that SQLite's test suite ("make test") does
catch the problem. In fact it gives up after 1000 errors.)
SQLite (and Fossil) is tested on Solaris/sparc64 using gcc 4.3.3. So
we do not see this problem.
It looks like compiling with LONGDOUBLE_TYPE set to "double" will
workaround the problem, but I haven't tried it yet.
Can you confirm that setting LONGDOUBLE_TYPE to "double" fixes the
problem. If it does, then we will
investigate fixing the problem in a configure script test that sets
LONGDOUBLE_TYPE appropriately.
Confirmed. Setting LONGDOUBLE_TYPE to "double" works around the problem.
The long-double-bug branch contains a proposed configure time test for
the bug and test suite entry to check for the bug. The configure time
test only runs on openbsd/sparc64 to minimize changes for other
platforms; the test suite item runs everywhere in case the bug also
exists on other platforms.
The cc-run.tcl file allows the configure script to run code snippets at
configure time. It looks like a lot of new code but it isn't really --
it is basically a lightly modified version of the cctest procedure from
cc.tcl.
The configure time test requires the "long long" data type, which is not
C89. If desired I could modify the test to first check for "long long"
support, but I think assuming "long long" support is reasonable given
that the test restricts itself to running on openbsd/sparc64.
--
Edward Berner
_______________________________________________
fossil-users mailing list
[email protected]
http://lists.fossil-scm.org:8080/cgi-bin/mailman/listinfo/fossil-users
