Re: [RFC] Add bad-branch-first option for git-bisect

On Mon, Jan 24, 2011 at 1:53 AM, Christian Couder <christian.couder@gmail.com> wrote:

Just some notes on gitbuilder's algorithm, since I haven't spent the time to fully understand Shuang's proposal.

I do understand at least one of his concerns, that is, that people like to do a lot of "messy" development on a branch, and when the branch is done, merge the whole messy branch into the "mainline". The messy branch would then have a lot of commits that break a lot of things before fixing them again later.

In a corporate environment, this method allows people to work all day, make frequent commits, pull from other branches at will, and never risk their lives by doing poorly-educated rebases. It works pretty well *until* you try to bisect, at which time all these messy commits start to bite you.

gitbuilder's bisection is a total hack around this situation, although it happens to work perfectly in the workflow it was designed for, thus making me feel clever.

Basically, we push/fetch *all* the branches from *everybody* into a
single repo, and build all of them as frequently as we can.  If you
think about it, if you have all the branches that someone might have
pulled/merged from, then you don't have to think of the git history as
a whole complicated DAG; you can just think of it as a whole bunch of
separate chunks of linear history.  Moreover, as long as people are
careful to only pull from a branch when that branch is passing all
tests - which you can easily see by looking at the gitbuilder console
- then playing inside each of these chunks of linear history can help
you figure out where particular bugs were introduced during "messy"
branches.

It also allows you a nice separation of concerns. The owner of the mainline branch (the "integration manager" person) only really cares about which branch they merged that caused a problem, because that person doesn't want to fix bugs, he/she simply wants to know who owns the failing branch, so that person can fix *their* bug and their branch will merge without breaking things.

So this is why gitbuilder uses "git rev-list --first-parent" during its "fake bisection" operation: because a different person is responsible for each "linear chunk" of history.

Note that you have to use --no-ff when merging if you want this to work reliably. But the build manager person can just remember to do that. Combining --no-ff and --ff-only (which sound mutually exclusive but aren't) is a way to be extra specially sure.

Now, if you aren't using gitbuilder, what we want from "bisection" is not quite the same, but let's imagine that you at least have a similar setup, where people *only* ever merge into the mainline by using --no-ff. In that case, you'd like a bisect operation that *starts* by using --first-parent, which will tell you which merge caused the problem. After that, you might want to bisect into the branch.

(I don't actually remember if 'git bisect' understands --first-parent correctly. gitbuilder doesn't exactly bisect either, but that's another story and not relevant right now.)

I can actually imagine that there are many more projects that do what I'm talking about - "messy" branches that get broken and fixed over time, then merge into a "clean" mainline - than projects (like the kernel and git.git) that try to keep all branches clean at all times. Thus, I could see some argument that a "--first-parents first" bisection would actually help out a lot of people, and maybe even deserves to be the default.

I don't really care though, I just use gitbuilder :)

Have fun,

Avery