Re: Fully automated bisecting with "git bisect run" [LWN.net]

[Darragh Bailey]

On Wed, Feb 11, 2009 at 10:42:14AM +0100, Uwe Bonnes wrote:
> > > > > > "Ben" == Ben Klein  writes:
> Ben> 2009/2/11 nn <[e-mail]>:
> >> Fully automated bisecting with "git bisect run" [LWN.net]:
> >> http://lwn.net/SubscriberLink/317154/5dec0c8146e58b61/
> >> 
> >> Would this be useful to add to the instructions on the wine wiki.
> >> 
> Ben> If I understand this correctly, this is only useful when the
> Ben> regression is something that can be tested without human
> Ben> interaction.  This makes it virtually useless for Wine, as most
> Ben> regressions involve loss of functionality some application.
>
> Ben> However, it could theoretically be useful for regressions that
> Ben> cause a complete crash of the application early on. Would it be any
> Ben> faster?  Probably not. Fewer commands to type, since you wouldn't
> Ben> need to run "git bisect bad" or "git bisect good" every time you
> Ben> test the app.
>
> It should work it there is a test in the test suite too

Comments to the article suggest that it gets significantly more complex
if you have 'toxic' patches in the tree. i.e. a patch that breaks your
test but is a different bug. You test needs to catch this and return a
different code to tell git to skip that revision.