Amit Kapila <amit.kapil...@gmail.com> writes:
> On Tue, Aug 17, 2021 at 6:40 AM Peter Smith <smithpb2...@gmail.com> wrote:
>> On Mon, Aug 16, 2021 at 11:31 PM Tom Lane <t...@sss.pgh.pa.us> wrote:
>>> Abstractly it'd be
>>>
>>> createpub := CREATE PUBLICATION pubname FOR cpitem [, ... ] [ WITH ... ]
>>>
>>> cpitem := ALL TABLES |
>>> TABLE name |
>>> ALL TABLES IN SCHEMA name |
>>> ALL SEQUENCES |
>>> SEQUENCE name |
>>> ALL SEQUENCES IN SCHEMA name |
>>> name
>>>
>>> The grammar output would need some post-analysis to attribute the
>>> right type to bare "name" items, but that doesn't seem difficult.
>> That last bare "name" cpitem. looks like it would permit the following
>> syntax:
>> CREATE PUBLICATION pub FOR a,b,c;
>> Was that intentional?
> I think so.
I had supposed that we could throw an error at the post-processing stage,
or alternatively default to assuming that such names are tables.
Now you could instead make the grammar work like
cpitem := ALL TABLES |
TABLE name [, ...] |
ALL TABLES IN SCHEMA name [, ...] |
ALL SEQUENCES |
SEQUENCE name [, ...] |
ALL SEQUENCES IN SCHEMA name [, ...]
which would result in a two-level-list data structure. I'm not sure
that this is better, as any sort of mistake would result in a very
uninformative generic "syntax error" from Bison. Errors out of a
post-processing stage could be more specific than that.
(Perhaps, though, we should *document* it like the latter way,
even if the actual implementation is more like the first way.)
regards, tom lane
