The filename never makes it into the database -- loadfromfile loads only the contents of the file
(and in any case, the file could be named pretty much anything anywhere, and it's never given us
any trouble for that.) After backup/restore, the file uploaded was the same one that wouldn't
upload previously -- data hadn't changed, only the database. It's not completely random, as it
seems like it mostly happens after we run a delete statement on rows in that table. It may simply
reveal a problem that's been there for a while, but it certainly at least seems somewhat
correlated. (It doesn't always happen though, only once every few (3 or 4) months. We usually
delete old version of our software after we get a dozen or so in the table, and we release often.)
We also detect the blob size as being zero bytes, not an intermediate value. It's always all or
nothing, but that could be because of some internal mechanism for detecting errors. (However, I
would expect it to tell me!)

-Philip

---------- Original Message -----------

>
> I've never stored executables in blobs so have no personal experience with
> it. But I have had many years with IB and although "random" does appear to
> be an occurence at first investigation, the final solution has revealed
> non-random reasoning. i.e. IB has never been known to me or any of my
> colleagues as able to produce randam behaviour. So there must be a reason
> somewhere. Binary blobs types are the first thing off the list... I imagine
> you have thoroughly debuggged your code? There's no chance the file name is
> dependent on something strange or missing path or assumption of current dir
> and no full path etc etc...You have ensured that your loadfromfile function
> is doing what's expected?
> if all this is aOK, then I would next suspect boundary markers.
>
> Alan
>

------- End of Original Message -------