[ppa-dev] What to do about bytea bug...

[Robert T. <ro...@xz...>]

I spoke with ioguix a few days ago on irc about releasing a new
version of PPA, and that I had hoped to take another crack at solving
the bytea problem. For those not familiar with it, the quick summary
is that if you attempt to insert or edit bytea data via PPA, it will
modify the data so that essentially you end up with corrupted data.
Unfortunately, the way PPA works, just editing another field of a
bytea row will also cause PPA to also update the bytea column, as we
update all values for a given row when any value is updated.

Now, I've tried a few different methods of trying to fix this, but
could come up with no working set of escape and unescape options that
would prevent this. (And as a note, it's made trickier because you
have to support multiple encoding formats, though I came up with
workarounds for that). At this point, I'm somewhat of the opinion it
can't be done. Now, I'd be happy for someone else to take a look and
show me the error of my ways, but barring that, I've spent the last
few days trying to come up with a work-around.

My initial idea was to just not allow editing of bytea data via PPA,
but I've run into some problems there as well, for example, if you
don't allow inserts of bytea data, then anyone who has a not null
bytea column would never be able to insert row data into their table.
That doesn't really seem like a good option.

So, barring someone figuring out a way to make this actually work, we
need to come up with a better work-around, or I guess just release the
next version with his bug in place. FWIW, this bug is in the current
stable version of PPA, so we're not making things worse, but it's kind
of frightening to know we have a subtle data corrupting behavior in
PPA.

Thoughts?

--
Robert Treat
conjecture: xzilla.net
consulting: omniti.com