failure doing massive revert

failure doing massive revert

[Mike Galbraith]

Greetings,

For reasons I'd rather not go into, I decided to create a merge free
tree to try to bisect.  I did this yesterday for a smaller range, and it
worked fine, and I was able to revert the reverts to re-apply.  Trying
to revert everything from v2.6.26..today croaked.

for i in `git rev-list --no-merges v2.6.26..HEAD`; do git revert $i < /dev/null; done

Got this far...

Author: Mike Galbraith <efault@gmx.de>  2008-09-18 10:50:58
Committer: Mike Galbraith <efault@gmx.de>  2008-09-18 10:50:58
Parent: 6753354a5984745b0121f7853e4e7a392e25adc7 (Revert "[ARM] 5185/1: Fix spi num_chipselect for lubbock")
Child:  0000000000000000000000000000000000000000 (Local uncommitted changes, not checked in to index)
Branch: master
Follows: v2.6.27-rc6
Precedes: 

    Revert "pktgen: multiqueue etc."
    
    This reverts commit e6fce5b916cd7f7f79b2b3e53ba74bbfc1d7cf8b.

...then began spewing fatal: Dirty index: cannot revert.  Probably me
being git-ignorant, but figured I'd mention it just in case.  I started
by whacking all source, followed by checkout -f master, so have no idea
what git means by local changes.

	-Mike

Re: failure doing massive revert

[Avery Pennarun]

On Thu, Sep 18, 2008 at 5:09 AM, Mike Galbraith <efault@gmx.de> wrote:
> For reasons I'd rather not go into, I decided to create a merge free
> tree to try to bisect.  I did this yesterday for a smaller range, and it
> worked fine, and I was able to revert the reverts to re-apply.  Trying
> to revert everything from v2.6.26..today croaked.
>
> for i in `git rev-list --no-merges v2.6.26..HEAD`; do git revert $i < /dev/null; done

Hmm, I don't think you can revert every single patch on a merged tree
that way for the same reason you can't just rebase it: history wasn't
linear.

I think something involving 'git rev-list --first-parent' and some
variant of "git diff $i $i^ | git apply" might work better, as it
would inherently squash merge commits, thus making them linearly
reversible (although throwing away large parts of history).

Throwing away history might not be what you want, but then again,
maybe it is.  It's the only way I know of to 100% reliably linearize
the history, anyway.

Also, if you use "&& done" instead of "; done" then it'll abort right
away when it has a problem.

Have fun,

Avery

Re: failure doing massive revert

[Mike Galbraith]

On Thu, 2008-09-18 at 14:26 -0400, Avery Pennarun wrote:
> On Thu, Sep 18, 2008 at 5:09 AM, Mike Galbraith <efault@gmx.de> wrote:
> > For reasons I'd rather not go into, I decided to create a merge free
> > tree to try to bisect.  I did this yesterday for a smaller range, and it
> > worked fine, and I was able to revert the reverts to re-apply.  Trying
> > to revert everything from v2.6.26..today croaked.
> >
> > for i in `git rev-list --no-merges v2.6.26..HEAD`; do git revert $i < /dev/null; done
> 
> Hmm, I don't think you can revert every single patch on a merged tree
> that way for the same reason you can't just rebase it: history wasn't
> linear.

Ok, nothing is broke, I just tripped over my white git-fu belt.

> I think something involving 'git rev-list --first-parent' and some
> variant of "git diff $i $i^ | git apply" might work better, as it
> would inherently squash merge commits, thus making them linearly
> reversible (although throwing away large parts of history).
> 
> Throwing away history might not be what you want, but then again,
> maybe it is.  It's the only way I know of to 100% reliably linearize
> the history, anyway.

Thanks, I'll save this reply in case I decide to resume hunt.

	-Mike