On Thu, 2015-09-03 at 11:45 +0100, Ian Jackson wrote: > sg-report-flight when testing X' (with a baseline of X) can justify a > failure of T(X',Y,Z) with a bisection failure of T(X,Y'',Z). > > If Y'' breaks T then this makes it look to sg-report-flight like T was > already broken in X; cr-daily-branch could then push X' even though it > is actually broken. > > This happened rarely, because cr-daily-branch's sg-report-flight would > only look at flights on the right branch, so only a bisection of T on > that branch can cause this, but nevertheless this can produce bad > pushes. > > So: have cr-daily-branch pass a --blessings option to cr-daily-branch, > so that it only looks at (usually) `real' rather than the default of > `real' and also `real-bisect'. > > An alternative, more complicated, approach would be for > sg-report-flight to compare versions of Y, Z, et al, when looking for > justifications, but I'm not sure this is desirable because it would > effectively reset the heisenbug compensator each time any other tree > changed. > > Signed-off-by: Ian Jackson <ian.jack...@eu.citrix.com>
Acked-by: Ian Campbell <ian.campb...@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
