On 31/03/2025 17:09, Tom Lane wrote:
Fabien COELHO <coelho@cri.ensmp.fr> writes:
[...] The attached patch allows \copy to use variable's values in place of table and file names:
Hm ... I'm on board with the general idea of the feature, but I find
this implementation quite horrid.
Indeed, I just added the stuff in the already quite peculiar manual lexer/parser for \copy. I did not think of addressing the why it is like that issue and try to fix it :-)
I would rather see us adjust the
lexing rules in psqlscanslash.l so that variable expansion happens
there when collecting \copy arguments. This would eliminate at
least part of the distinction between OT_WHOLE_LINE and OT_NORMAL
modes, and we'd have to have some discussion about how far to go
there. Or maybe just change exec_command_copy to use OT_NORMAL
not OT_WHOLE_LINE? If we modify the behavior of OT_WHOLE_LINE
then the ability to expand variables would start to apply in the
other commands using that, notably \!. I think that's at least
potentially good, but perhaps the blast radius of such a change
is too large.
I'm not sure that such \copy salvage to using lex is easy because:
(1) it seems that is the only command which is really full SQL hidden in a backslash command
(2) on one line without requiring a final ';',
(3) the client needs to actually parse it and modify it to some degree before sending it to the server.
so the implication for trying to maintain compatibility without adding weirdness seem slim.
Anyway, my feeling about it is that \copy parsing is a huge hack
right now,
Yes.
and I'd rather see it become less of a hack, that is
more like other psql commands, instead of getting even hackier.
I think that I'll have a try with Corey suggestion to extend COPY rather than change \copy.
--
Fabien.
