Обсуждение: pg_dump -s -b

All,

While reviewing and considering the patch to add the --no-blobs option
to pg_dump, which I entirely agree with, I came across our claim that
passing -b to pg_dump would cause blobs to be output, regardless of
other options set.

To wit, our docs specifically say:

----
Include large objects in the dump. This is the default behavior except
when --schema, --table, or --schema-only is specified, so the -b switch
is only useful to add large objects to selective dumps.
----

However, a 'pg_dump -s -b' won't include blobs, which appears to be due
to the masking performed in _tocEntryRequired() whereby we strip the
REQ_DATA out if we are in schema-only mode.

Further, though perhaps less surprising, we don't include blobs when
--section=post-data is set, and only the definition of the blob (but no
data) when --section=pre-data is used.

This appears to go back at least as far as all currently supported
branches.

We have a few options here.  The documentation says 'selective dumps',
which we could infer to mean 'dumps where the user specified a table' or
similar and exclude 'schema-only' from that case.  I don't believe this
was the original intent, given this comment in pg_dump.c:

/** Dumping blobs is now default unless we saw an inclusion switch or -s* ... but even if we did see one of these, -b
turnsit back on.*/
 

Given the lack of field complaints about this, I'm pretty tempted to
just adjust that comment and the documentation.  Alternativly, we could
hack things up in a simimlar manner to what was done in a7e5457, and
cause blobs to be dumped even in schema-only mode if -b is passed.

Thoughts?

Thanks!

Stephen

Stephen Frost <sfrost@snowman.net> writes:
> To wit, our docs specifically say:
> ----
> Include large objects in the dump. This is the default behavior except
> when --schema, --table, or --schema-only is specified, so the -b switch
> is only useful to add large objects to selective dumps.
> ----

Clearly, this is an oversimplification.

> However, a 'pg_dump -s -b' won't include blobs, which appears to be due
> to the masking performed in _tocEntryRequired() whereby we strip the
> REQ_DATA out if we are in schema-only mode.
> Further, though perhaps less surprising, we don't include blobs when
> --section=post-data is set, and only the definition of the blob (but no
> data) when --section=pre-data is used.

IMO, blobs are data, so those behaviors are correct.  The docs should
be fixed to match the code.
        regards, tom lane

* Tom Lane (tgl@sss.pgh.pa.us) wrote:
> IMO, blobs are data, so those behaviors are correct.  The docs should
> be fixed to match the code.

Works for me.

Thoughts on the attached?

Thanks!

Stephen