This patch seems extremely messy to me. Unless I'm missing something,
-w just plain fails except when you are dumping a specific table (ie,
-t must be given as well). And heaven help you if you specify a
different table in -t than the one -w is selecting from. This isn't
well thought out.
I'm not at all convinced that such a thing belongs in pg_dump anyway.
It'd be more useful as a manually-invokable feature, I think. You
can almost do this in psql withselect * from table where something\g outfile
but I don't think you can get psql to emit the data in a form that can
be reloaded reliably (it won't quote data characters that look the same
as column delimiters, for instance).
What would seem to make sense is adding a WHERE-clause option to
COPY TO, and then you could goCOPY table TO 'myfile' WHERE ...
We already have column-list support in COPY, so we can already slice the
table vertically --- WHERE would let you slice it horizontally, which
seems a natural extension. (BTW, has anyone taught psql's \copy about
column lists? AFAIR the original patch was only against the backend.)
I'm finding it hard to visualize situations where I'd want the extra
baggage of pg_dump for something like this. If I want the schema at
all, I'll probably want it separate from the data so that I can hack
the schema conveniently --- so I'd want to do a "pg_dump -s -t table"
and then do the selective copying separately.
regards, tom lane
