On 7 Apr 2016, at 11:18 AM, Ben Hutchings <b...@decadent.org.uk> wrote: > On Thu, 2016-04-07 at 01:02 +0000, Potter, Tim (HPE Linux Support) > wrote: >> On 7 Apr 2016, at 10:52 AM, Ben Hutchings <b...@decadent.org.uk> >> wrote: >>> Given the low quality and lack of unit tests in many scientific >>> applications, how confident can we be that the 'old' packages (that >>> have now built with newer toolchains and libraries) actually still >>> produce the same results they used to? If we are not, even that >>> historic value is lost. >> >> Full archive rebuilds are done every so often. The switchover to the >> gcc-5 toolchain >> was an example and everything was rebuilt at least once during that >> time. My understanding >> is that packages are dropped if they don't build in this case, and >> no-one steps up to fix them >> within a reasonable (months) period of time. > > You are missing the point, which is that while they still build with > the new toolchain (possibly after a developer without intimate > knowledge of the program makes a best-effort fix) we don't know that > they behave the same way.
OK - good point. I wonder if there is any information about how many packages run unit tests? It would be interesting to see the data. I get the impression that more upstream packages have built-in tests that can be run as part of dpkg-buildpackage (e.g Python, Perl, Ruby, Java and Go) but maybe that's just because I've been working in those environments recently. Tim.
signature.asc
Description: Message signed with OpenPGP using GPGMail