Re: [jgit-dev] Re: Gerrit offline for JGit rewrite

["Shawn O. Pearce" <spearce@xxxxxxxxxxx>]

Robin Rosenberg <robin.rosenberg@xxxxxxxxxx> wrote:
> onsdagen den 10 mars 2010 16.28.31 skrev  Shawn O. Pearce:
> > > Inline comments on my last pushed change (#315) have disappeared.
> > > In the patch set overview it is correctly reported how many inline
> > > comments there are for each file. However, If I open any of the files the
> > > inline comments are not shown.
> > 
> > Frell.
> > 
> > 82 inline comments in 13 changes had their line numbers set to 0 when
> > I attempted to merge in the delta of the header impact.  Yesterday I
> > also pulled the brillant move of wiping the backup file.  Arrrrgh.
> > 
> > I've moved them all to line number 1, so they appear in the file, but
> > they are at the wrong spot.  Sorry.  Without digging through email
> > archives and manually matching up these 82 comments back to their
> > original line numbers, there's no way to reposition these accurately.
> 
> I wouldn't go that far. It would be possible to automatically detect the
> repositioning since you have the file before and after and the diff would
> tell us what moved and how much. It's not worth it in general, but then
> with Gerrit at Eclipse, it might be useful as a standard feature. I think your
> compromise is ok this time.

There were 584 inline comments in JGit last night.  Of those,
all but 82 got repositioned using the algorithm you describe.

The 82 in question repositioned to 0, due to an issue with the
update that I just don't care to identify.  Its likely something
about the diff stat between the old commit and the new commit
didn't report the delta as expected when the file was a new file,
and we somehow caused the line number to truncate to 0.  No clue
why that would happen though.

It probably wasn't worth the effort.  MySQL was a real PITA because
it wanted an exclusive table lock in order to perform the scan
and update.  And it wasn't very fast about the 4 way table join.
But I did try.

-- 
Shawn.