Le 02/06/2013 11:26, Konstantin Tokarev a écrit : > > 02.06.2013, 01:36, "Sylvestre Ledru" <[email protected]>: >> On 01/06/2013 17:55, Konstantin Tokarev wrote: >> >>> 01.06.2013, 19:39, "Sylvestre Ledru" <[email protected]>: >>>> http://buildd-clang.debian.net/coverage/ >>>> Your patch fixed the issue! >>>> Thanks to it, the LLVM toolchain is close to the 80 % lines test >>>> coverage (which is excellent). >>> Note that line-based coverage is far from being precise. >> Could you describe your criticism a bit more ? > You can read through review of code coverage metrics in [1]. > > In short, if you have a code line > > if ((a<b) && (c<d)) doIt(); > > you will get 100% line coverage for it if condition is ever evaluated, > but you'll get 100% condition/decision coverage only if tests cover 4 > different situations: > > 1) a<b is true, c<d is true > 2) a<b is false, c<d is true > 3) a<b is true, c<d is false > 4) a<b is false, c<d is false Yes, right but the current solution remains interesting (even if it is not perfect). >>> To get more realistic estimation of coverage, you need to use coverage >>> tool that >>> provides decision and condition coverage, e.g. TestCocoon. >> Are you aware of any free software providing this feature ? I would be >> happy to try. > I'm using TestCocoon [2]. It is open source tool under GPLv3 license. > Unfortunately, > it's effectively dead right now, because it was bought by Froglogic and is > developed > as proprietary tool Squish Coco [3] now. On the positive side, Squish Coco is > available > for free (as beer) for non-commercial usage [4]. > Yeh but I would prefer to use FLOSS software...
Sylvestre _______________________________________________ lldb-dev mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/lldb-dev
