Re: [Gnu-arch-users] [BUG] pika escaping corrupts taglines containing "

[Colin Walters]

On Wed, 2004-09-15 at 02:22 +0200, address@hidden wrote:
> > Hi,
> > 
> > I have a project that uses taglines, and one of the taglines contains
> > the " character.  Before pika escaping, this character was stored
> > literally everywhere.  With pika escaping, it is stored in changesets
> > as\". This obviously breaks compatibility with older versions of tla,
> > and can cause silent corruption.
> 
> I shortly investigated the cause, it seems (only 95% sure so far) that
> tla does str_cmp on ids without unescaping them. thats seems to be a
> real bug. 

Note that older clients don't know how to unescape.  Thus they won't be
able to read escaped changesets, even if we change newer versions to do
unescaping before str_cmp.

But as we discussed on IRC I think at this point we should make a clean
break and mark newer changesets with a version number (say 1, assume the
old version was 0).

If we are going to go this route and break older clients, I really want
it to be obviously broken (i.e. stop with an error), not the extremely
subtle breakage that leads to archive corruption that can happen now.

Perhaps we should modify new clients to update the .archive-version file
on any archive transaction if it's format version 2, and change it to 3?
Ugly, but I don't see a better solution...