Mismerge at r330897 in stable/11, Audit report

[Glen Barber]

On Wed, Mar 28, 2018 at 07:49:06PM -0700, Rodney W. Grimes wrote:
> > On Wed, Mar 28, 2018 at 07:17:20PM -0700, Eitan Adler wrote:
> > > On 28 March 2018 at 19:04, Rodney W. Grimes
> > > <freebsd at pdx.rh.cn85.dnsmgr.net> wrote:
> > > >> On 28 March 2018 at 18:35, Rodney W. Grimes
> > > >> <freebsd at pdx.rh.cn85.dnsmgr.net> wrote:
> > > >> >> >> Hi!
> > > >> >> >>
> > > >> >> >> This part of the MFC is wrong:
> > > >> >> >>
> > > >> >> >> https://svnweb.freebsd.org/base/stable/11/sys/sys/random.h?limit_changes=0&r1=330897&r2=330896&pathrev=330897
> > > >> >
> > > >> > Can we try to identify exactly what rXXXXXX that is a merge of?
> > > >> >
> > > >> >> >> Could you please MFC back the other random related changes too? Some
> > > >> >> >> of them made by cem at .
> > > >> >> >>
> > > >> >> >> On 3/14/18, Eitan Adler <eadler at freebsd.org> wrote:
> > > >> >> >>> Author: eadler
> > > >> >> >>> Date: Wed Mar 14 03:19:51 2018
> > > >> >> >>> New Revision: 330897
> > > >> >> >>> URL: https://svnweb.freebsd.org/changeset/base/330897
> > > >> >> >>>
> > > >> >> >>> Log:
> > > >> >> >>>   Partial merge of the SPDX changes
> > > >> >> >>>
> > > >> >> >>>   These changes are incomplete but are making it difficult
> > > >> >> >>>   to determine what other changes can/should be merged.
> > > >> >> >>>
> > > >> >> >>>   No objections from:        pfg
> > > >> >> >>>
> > > >> >> > Am I missing something? If this MFC was supposed to be of the SPDX
> > > >> >> > license tagging, why does it have any functional changes?
> > > >> >> >
> > > >> >> > Especially changes to random(4)?
> > > >> >>
> > > >> >> This was my failure. I only spot checked & compile-checked the diff
> > > >> >> since I expected all changes to be comments/SPDX.
> > > >> >>
> > > >> >> However, I must have gotten carried away and included a few too many
> > > >> >> revisions. Unfortunately some people have already merged fixes to my
> > > >> >> failure and thus this can't be reverted as is without also reverting
> > > >> >> those fixes.
> > > >> >>
> > > >> >> That said, I should do that since this commit message is utterly wrong.
> > > >> >
> > > >> > We do not have to revert r330897, with what follows I think
> > > >> > we can easily find the revisions to revert from stable/11.
> > > >> > ...
> > > >>
> > > >> While we don't have to revert it I'd rather do so than have bogus history.
> > > >
> > > > Reverting wont remove that history, thats a one way deal,
> > > > and I think if we revert the bogus merges with the wrong
> > > > history thats as good as its gona get.
> > > >
> > > >>
> > > >> >From a look it seems the following was also merged:
> > > >> r316370, r317095, r324394, and a few others.
> > > >>
> > > >> Is there a reason you don't want me to revert the changes?
> > > >
> > > > Repository churn is my main concern.
> > > >
> > > > It touches 6000+ files some of which have probably
> > > > been touched since.   A very carefull pre commit
> > > > audit would need to be done.
> > > >
> > > > Then another commit to 6000+ files to put it back,
> > > > also needing a pre-commit audit. (Pretty easy now
> > > > that I have a filter.)
> > > 
> > > I'm actually using the same filter you pasted above to verify that my
> > > changes are only reverting said files. That said, while I'd prefer to
> > > revert, I'll defer to others if they have a differing opinion.
> > > 
> > > 
> > > Note that I won't have access my dev box after tomorrow for about a week.
> > > 
> > 
> > IMHO, if you are going to be away for over a week while we're headed
> > directly into the 11.2 release cycle, revert the change.  What you
> > committed is not what was intended, clearly, and the commit message does
> > not reflect what had happened (as you noted).
> > 
> > Any disagreements on this decision should be directed to me specifically
> > in this case.
> 
> Glen,
> 	I would rather not revert, as I believe that would cause more
> damages as people have already cleaned up some of the mis merge from
> this commit.  I am pretty sure a revert would lead to a broken tree.
> 
> In Eitans absence I am willing to take responsiblity to untangle
> the wrong bits and clean up stable/11.
> 
> Ok?
> 

I disagree with this approach.  A mishandled merge which as in this case
contains extraneous changes limit our ability to properly audit the
branch.

My strong feeling is that if "too many" revisions were merged, we should
not piecemeal break out what should not have been included.  We should,
instead, revert the offending commit as an "oops", redo the intended
merge, and then in a separate commit re-apply the bits from the
offending merge in a completely separate commit.

While this may be perceived as churn against the tree, it keeps tracking
what happened much easier to manage, especially if we were to have to
rely on mergeinfo.

> Eitan,
> 	Are you ok with that as well?
> 

Glen

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/svn-src-all/attachments/20180329/d5c3eb1b/attachment.sig>