Jim Starkey said:

>For the record, I don't think blob segments have proven their worth.
>Can anybody think of any good reason to keep them part of the
>architecture?

What he did not say is that there are two different types of blob,
segmented and stream. The segmented blobs are stored with a binary
segment length. You can store a segment of any length, but when
you retrieve it, the engine will never give you more than one
segment at a time. Stream blobs are stored as a stream of bytes.
When you store into one, you send some number of bytes. When you
retrieve data from a stream blob, you get the smaller of the
remaining bytes in the blob or a full buffer.

No one has ever seen a stream blob (except fancied up
as an array) because the documentation group was swamped with new
features and decided to cut some. (V3) If stream blobs were
available and documented, nobody would use segmented blobs

People, including me, used text segmented blobs in two ways.
Since each line was a segment, I often saw no reason to store
the line terminator. But not always.

Nobody would be that dumb today, but the fact remains that it is
possible to infer semantic content from segment length. Which is
a reason to leave them around, but push stream blobs instead.

Ann