Re: proposal: multiple psql option -c - Mailing list pgsql-hackers

From Pavel Stehule
Subject Re: proposal: multiple psql option -c
Date
Msg-id CAFj8pRCaE_kxwguohbQmODA-JGKnZRRa3y41GdJV3brsix-0TA@mail.gmail.com
Whole thread Raw
In response to Re: proposal: multiple psql option -c  (Robert Haas <robertmhaas@gmail.com>)
List pgsql-hackers


2015-11-17 21:00 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:
On Tue, Nov 17, 2015 at 2:25 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Mon, Nov 16, 2015 at 6:05 PM, Andrew Dunstan <andrew@dunslane.net> wrote:
>>> I honestly don't see what's so confusing about it, and if there is any
>>> confusion then surely we could make sure what's happening is well
>>> documented.
>
>> +1.  I'm actually kind of wondering if we should just back up and
>> change the way -c works instead, and allow it to be specified more
>> than once.  The current behavior is essentially a crock that has only
>> backward compatibility to recommend it, and not having two
>> confusingly-similar options is of some non-trivial value.
>
> Well, it's not *entirely* true that it has only backwards compatibility
> to recommend it: without -c in its current form, there would be no way
> to test multiple-commands-in-one-PQexec cases without hacking up some
> custom test infrastructure.  That's not a very strong reason maybe, but
> it's a reason.

True.  We could have a --no-split-commands option for that case, perhaps.

> And backwards compatibility is usually a strong argument
> around here anyway.

Yeah, but not - at least in my book - at the expense of being stuck
with a confusing interface forever.

> I've not been following this thread in any detail, but have we considered
> the approach of allowing multiple -c and saying that each -c is a separate
> PQexec (or backslash command)?  So the semantics of one -c wouldn't change,
> but commands submitted through multiple -c switches would behave
> relatively unsurprisingly, and we wouldn't need two kinds of switch.
>
> Another issue here is how -1 ought to interact with multiple -c.

On a code level, I think the issue here is that ACT_SINGLE_QUERY
bypasses a lot of stuff that happens in the ACT_FILE case: directly in
main, there's process_psqlrc(); inside process_file(), there's the
single_txn handling; then inside MainLoop, there's splitting of
commands and cancel handling and various other stuff.  In my
imagination, it's like this because originally psql wasn't nearly as
complicated as it is now, and as features got added in various places,
-c gradually diverged.  I don't know whether that's really what
happened, but it seems to me that it would be good to bring those
things back together.

A few years ago there was a proposal to not only allow multiple -c
options, but to allow -c and -f to be intermingled.  This seems really
rather useful; I'd like to be able to type psql -c do_this_first -f
script.sql and have that work.  But of course it's kind of hard to
figure out how that should behave given the various differences
between -c and -f.  I think in the long run we'll be better off
rationalizing the interface; I really doubt how many people, even on
this mailing list, can even enumerate all the differences between -c
and -f.  If it's too complicated for hackers to remember, it's
probably not very good for users either.

This shouldn't be hard. With proposed patch, you can do "-C '\i xxx' -C '\i yyyy'" - that is effectively same like multiple -f

and this mix can be really useful }in any order).

Regards

Pavel
 

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

pgsql-hackers by date:

Previous
From: Robert Haas
Date:
Subject: Re: proposal: multiple psql option -c
Next
From: Robert Haas
Date:
Subject: Re: [DESIGN] ParallelAppend