Boost Testing :
Subject: Re: [Boost-testing] [rational] Updated for fix; run tests
From: Ben Pope (benpope81_at_[hidden])
Date: 2013-08-22 22:12:39
On 22/08/13 11:12, Daryle Walker wrote:
> Earlier today, I uploaded a fix a narrowing-conversion bug to boost::rational. (Ticket #5855, change-set ) But I did another check-in first (change-set ).
> The lower-numbered change represented a problem while fixing the ticket. C++11 adds two static members to std::numeric_limits. One of them, max_digits10, represents how many digits are needed to be printed to (near) guarantee text/binary conversion. Boost.Test uses it in its C++11 code, so it expects any version of numeric_limits to have it, which caused errors with Boost.Rational since that part of the test code hasn't been touched since it was written nearly 7 years ago! (There was a minor tweak over 5 years ago.)
> One advantage with Git is that I could have checked both changes separately. Here, for Subversion, I had to back out the second change, commit the first, then redo the second to get them recorded independently.
> I just did a quick look at our regression tests, and most of them involve one or both of the issues I just fixed. So can everyone do a test cycle sooner rather than later? (The "teeks99-08(e|f|g|h)-Ubnutu12.04-64" teams need to make sure Clang is installed and in the PATH.)
> Daryle W.
Trunk is looking good, but for some reason my runner (BP x86_64 C++11)
doesn't have that problem for gcc-4.8.1 on release, but it does have it
for clang 3.2 and 3.3, any ideas?