scary changes...

[Dirk Hohndel]

On Thu, 2013-11-14 at 13:37 +0200, Willem Ferguson wrote:
> I have driven this one around a few blocks and it did not misfire even once.

Thanks, Willem. You have been one of the most active and diligent
testers lately. MUCH appreciated.

/D


> On 14/11/2013 12:46, Dirk Hohndel wrote:
> >
> > Here's the commit message of the last commit I just pushed:
> >
> > commit ed41d5a74490b9af03ece57015d6867bbf00df14
> > Author: Dirk Hohndel <dirk at hohndel.org>
> > Date:   Thu Nov 14 19:36:41 2013 +0900
> >
> >      Sort the dive table after adding a dive
> >      
> >      This commit scares me. Any pointers to dives or indices into the dive
> >      table... anything like that is invalid after we resort the table. Well,
> >      not technically "invalid" (as in bad pointers), but after re-sorting the
> >      table these will possibly not be pointing at what we expected.
> >      
> >      This starts with the selection being "wrong" after we add a dive that
> >      isn't the last dive (once we click OK the chronologically last dive will
> >      be selected).
> >      
> >      But of course without doing this, our #1 assumption about the dive_table
> >      is broken. The dive_table is supposed to be in chronological order.
> >      
> >      Best advice of course would be "don't enter dives out of order" - but of
> >      course that's not realistic.
> >      
> >      Fixes #234
> >      
> >      Signed-off-by: Dirk Hohndel <dirk at hohndel.org>
> >
> >
> > I am reasonably certain that this is the right thing to do, but as the
> > commit message indicates, I'm rather worried about the consequences.
> > I have stared at this code for too long and would really appreciate some
> > others (especially Linus...) to take a look.
> >
> > We do make the assumption that our dive_table is sorted in various
> > places (that's how the bug report #234 pointed me to this... if you add
> > dives out of chronological order you get negative time intervals between
> > dives... oops). But sadly we also have the habit of keeping either dive
> > indices (things like selected_dive) or (far worse) pointers to dives
> > around in memory. We should NEVER do the latter (except for time periods
> > where we KNOW that the dive_table won't be modified). The former
> > is /usually/ ok, but will still end up causing trouble if we for example
> > merge two dives or (as in this case) add dives in the middle of the
> > dive_table.
> >
> > Please test.
> >
> > /D
> >
> > _______________________________________________
> > subsurface mailing list
> > subsurface at hohndel.org
> > http://lists.hohndel.org/cgi-bin/mailman/listinfo/subsurface
> >
>