Between a trusted commit

PHYSICS · Multi-part · Fermi University Physics · Computational Physics · Research software practice, verification and validation · Research software practice, verification and validation

Scientific Computing for Physicists, Ch A repository others can build on: collaboration, history and workflows

Problem

c_0

, where the orbit integrator's seeded energy-drift test passes, and the current commit

c_{64}

, where it fails, lie the commits

c_1,\dots,c_{64}

on a single line of history. Unknown to the group,

c_{10}

introduced a fault in the half-step velocity update,

c_{30}

repaired it, and

c_{50}

put it back. The test therefore fails on

c_{10}

c_{29}

and on

c_{50}

c_{64}

, and passes on every other commit. The group runs bisection search with

c_0

marked good and

c_{64}

marked bad, always testing

c_m

with

m=\lfloor(g+b)/2\rfloor

, where

c_g

is the current good end and

c_b

the current bad end. ((a)) List the commits tested, in order, with each verdict. State how many runs the search takes and which commit it names as the first bad one. ((b)) The named commit fails and its parent passes, exactly as the search promises. Explain why it is nonetheless not the commit that first introduced the fault, and name the assumption of the chapter's key result that this history breaks. ((c)) Reading the named commit's diff shows that it undoes the change made in

c_{30}

, whose message reads "repair half-step update". Use this to choose a new bad end for a second search from

c_0

. Give the most runs that search can need, and the commit it names.

Browse the problem library