Thread: PSQL commands: \quit_if, \quit_unless
This patch adds two very simple psql commands: \quit_if and \quit_unless.
Each takes an optional string parameter and evaluates it for truthiness via ParseVariableBool().
If a true-ish value is passed to \quit_if, psql will behave as if the user had input \quit.
\quit_unless will do nothing if the value given was true-ish, and will \quit in any other circumstances.
Examples below show the behavior:
psql (10devel)
Type "help" for help.
# \quit_if
# \quit_unless
$ psql postgres
psql (10devel)
Type "help" for help.
# \quit_if f
# \quit_if 0
# \quit_if false
# \quit_if 2
unrecognized value "2" for "\quit_if"; assuming "on"
$ psql postgres
psql (10devel)
Type "help" for help.
# \quit_unless 2
unrecognized value "2" for "\quit_unless"; assuming "on"
# \quit_unless f
$
Attachment
Hello Corey, > This patch adds two very simple psql commands: \quit_if and \quit_unless. A few comments about the feature design: I'm unsure about the name, esp with '_'. There are some \lo_* commands, but others rely on pasted words (\crosstabview, \errverbose, ...). I'm wondering if an simplistic interpreted \if \elsif/\else \fi would make more sense: Quitting seems a little bit definitive, and means that if I have some alternatives then I have to have something that relaunch another script outside... When \includes are process, does \quit stop the include or the full script. I'm afraid it is the script. Now probably an \if... would have also some drawbacks, but ISTM that there could be less of them. There is no test provided with the patch. -- Fabien.
2016-11-28 20:03 GMT+01:00 Fabien COELHO <coelho@cri.ensmp.fr>:
Hello Corey,This patch adds two very simple psql commands: \quit_if and \quit_unless.
A few comments about the feature design:
I'm unsure about the name, esp with '_'. There are some \lo_* commands, but others rely on pasted words (\crosstabview, \errverbose, ...).
There are not any other conditional statements - so using "_" can be better
I'm wondering if an simplistic interpreted \if \elsif/\else \fi would make more sense:
The \if \ese \elseif is not in contradiction with \quit_if - \if is more generic, \quit_if is simple for implementation, one liner, and best readable
Pavel
Quitting seems a little bit definitive, and means that if I have some alternatives then I have to have something that relaunch another script outside...
When \includes are process, does \quit stop the include or the full script. I'm afraid it is the script.
Now probably an \if... would have also some drawbacks, but ISTM that there could be less of them.
There is no test provided with the patch.
--
Fabien.
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
On Mon, Nov 28, 2016 at 2:03 PM, Fabien COELHO <coelho@cri.ensmp.fr> wrote:
Hello Corey,This patch adds two very simple psql commands: \quit_if and \quit_unless.
A few comments about the feature design:
I'm unsure about the name, esp with '_'. There are some \lo_* commands, but others rely on pasted words (\crosstabview, \errverbose, ...).
I'm completely flexible with regard to the names.
I'm wondering if an simplistic interpreted \if \elsif/\else \fi would make more sense:
The problem is that \if \elsif \else \fi is anything but simplistic, and would be a vastly bigger change. Detecting nested if-thens, detecting un-matched if-thens, etc.
Quitting seems a little bit definitive, and means that if I have some alternatives then I have to have something that relaunch another script outside...
When \includes are process, does \quit stop the include or the full script. I'm afraid it is the script.
It behaves exactly as if the script contained \quit at that location.
Now probably an \if... would have also some drawbacks, but ISTM that there could be less of them.
It'd be nice, but I'm trying to keep things simple.
There is no test provided with the patch.
Indeed, but testing such a feature is hard within our current test harness. I welcome suggestions for how to convert the example script in my first email to tests.
Corey Huinker <corey.huinker@gmail.com> writes:
> On Mon, Nov 28, 2016 at 2:03 PM, Fabien COELHO <coelho@cri.ensmp.fr> wrote:
>> I'm wondering if an simplistic interpreted \if \elsif/\else \fi would make
>> more sense:
> The problem is that \if \elsif \else \fi is anything but simplistic, and
> would be a vastly bigger change. Detecting nested if-thens, detecting
> un-matched if-thens, etc.
Yes, but... We've generally refrained from inventing any control flow
metacommands for psql, and TBH this does *not* seem like the place to
start. If we take this patch we are very likely to find that it doesn't
fit in at all whenever someone does show up with a proposal for control
flow features.
If we had an agreed-on sketch for what that feature set would look like,
I wouldn't necessarily object to implementing commands like these first.
But as it stands I think we'd be painting ourselves into a corner.
There's no reason to assume a-priori that this patch creates either naming
conventions or semantics (e.g. what is suitable as a boolean expression)
that we'd be happy with in a larger context.
As far as the original problem goes, I wonder whether what you really
want isn't a \quit command that lets you specify psql's exit code.
I'm not quite seeing how this proposal advances the problem of
communicating between psql and shell portions of a script; but I can
see how returning 0 (ok) vs 1 (error) vs 2 (some special case) would
help with that.
regards, tom lane
Corey Huinker <corey.huinker@gmail.com> writes:
> On Mon, Nov 28, 2016 at 2:03 PM, Fabien COELHO <coelho@cri.ensmp.fr> wrote:
There's no reason to assume a-priori that this patch creates either naming
conventions or semantics (e.g. what is suitable as a boolean expression)
that we'd be happy with in a larger context.
Would we be happy borrowing the definition of FOUND from pl/pgsql?
As far as the original problem goes, I wonder whether what you really
want isn't a \quit command that lets you specify psql's exit code.
Actually, I'm seeing this as basically an assertion capability and maybe should be named as such
\assert_is
\assert_isnot
David J.
As far as the original problem goes, I wonder whether what you really
want isn't a \quit command that lets you specify psql's exit code.
The ability to specify an exit code was part of the brainstorming, yes. But with it was the ability to conditionally quit.
Actually, I'm seeing this as basically an assertion capability and maybe should be named as such\assert_is\assert_isnot
That came up too! I see real value in the ability to test for error conditions. I just had a more immediate need for a non-error exit condition.
On 11/28/2016 04:07 PM, Tom Lane wrote: > ... We've generally refrained from inventing any control flow > metacommands for psql I think it's really time we seriously considered adding some flow control logic, though. I'm mildly tired of either jumping through hoops to get around the lack or having to switch to using some other thing that has the required logic mechanism (hello Perl). cheers andrew
Andrew Dunstan <andrew@dunslane.net> writes:
> On 11/28/2016 04:07 PM, Tom Lane wrote:
>> ... We've generally refrained from inventing any control flow
>> metacommands for psql
> I think it's really time we seriously considered adding some flow
> control logic, though.
Yeah, maybe. I'd be interested to see a fully worked out proposal
for that.
regards, tom lane
Hello, >> I think it's really time we seriously considered adding some flow >> control logic, though. > > Yeah, maybe. I'd be interested to see a fully worked out proposal > for that. I agree that designing a fuller proposal before including individual parts would be great and result in a more consistent result. In order to bootstrap the discussion, I suggest the following: - boolexpr is a simple "boolean" (t, non 0 int, non empty string.. as proposed by Corey and Pavel) or !/not boolexp ;it could be extended if necessary, but I would try to avoid that, as - actual more complex expressions could be left to the server through SQL which simplifies the client a lot by avoidingan expression language altogether - then having a conditional block is very versatile and can be adapted to many use cases... maybe all - \quit CODE, or I would prefer \exit CODE, could be used to exit while controlling the status It could look like (although I do not like gset in this context, but anyway): SELECT ... AS has_foo_extension \gset SELECT ... AS has_bla_extension \gset \if :has_foo_extension ... \elif :has_bla_extension ... \else -- no foo nor bla extension \echo please install foo or bla extension \exit 1 \fi -- extension... SELECT ... AS has_xxx_feature \gset \if ! :has_xxx_feature \echo "feature xxx is needed, aborting" \exit 2\fi ... -- Fabien
2016-11-29 8:44 GMT+01:00 Fabien COELHO <coelho@cri.ensmp.fr>:
Hello,I think it's really time we seriously considered adding some flow
control logic, though.
Yeah, maybe. I'd be interested to see a fully worked out proposal
for that.
I agree that designing a fuller proposal before including individual parts would be great and result in a more consistent result.
In order to bootstrap the discussion, I suggest the following:
- boolexpr is a simple "boolean" (t, non 0 int, non empty string.. as
proposed by Corey and Pavel) or !/not boolexp ; it could be extended if
necessary, but I would try to avoid that, as
Now, the psql statements are designed do nothing in syntax error. I am not sure about be more strict in this case. I see strong advantages - but it can be little bit different than current behave.
- actual more complex expressions could be left to the server through SQL
which simplifies the client a lot by avoiding an expression language
altogether
- then having a conditional block is very versatile and can be adapted to
many use cases... maybe all
- \quit CODE, or I would prefer \exit CODE, could be used to exit while
controlling the status
It could look like (although I do not like gset in this context, but anyway):
SELECT ... AS has_foo_extension \gset
SELECT ... AS has_bla_extension \gset
\if :has_foo_extension
...
\elif :has_bla_extension
...
\else -- no foo nor bla extension
\echo please install foo or bla extension
\exit 1
\fi -- extension
...
SELECT ... AS has_xxx_feature \gset
\if ! :has_xxx_feature
I prefer the commands instead symbols - the parsing and processing symbols should be more complex than it is now. A psql parser is very simple - and any complex syntax enforces lot of code.
\if_not
Regards
Pavel
\echo "feature xxx is needed, aborting"
\exit 2
\fi
...
--
Fabien
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
Hello Pavel, > Now, the psql statements are designed do nothing in syntax error. I am not > sure about be more strict in this case. I see strong advantages - but it > can be little bit different than current behave. Indeed, an error on a conditional construct should stop the script, which is slightly different that "go on whatever", which is okay in the interactive mode. I do not see that as an issue, just as features which are more interactive vs script oriented and behave accordingly: typing a \if in interactive does not makes much sense, because you have tested something is there, so you know the answer and can act directly, so there is no point using a condition. >> \if ! :has_xxx_feature > > I prefer the commands instead symbols - the parsing and processing > symbols should be more complex than it is now. A psql parser is very > simple - and any complex syntax enforces lot of code. \if_not My 0,02 €, which is just a personal opinion: I think that handling manually "!/not" would be worth the effort rather than having two commands, especially if the boolean expression syntax may be extended some day and the negative if would become obsolete. If there is a negative condition syntax, I would slightly prefer \ifnot to \if_not or worse \unless. I would disaprove strongly of \unless because it looses the clear symmetry with a closing \fi. -- Fabien.
My 0,02 €, which is just a personal opinion:
I think that handling manually "!/not" would be worth the effort rather than having two commands, especially if the boolean expression syntax may be extended some day and the negative if would become obsolete.
If there is a negative condition syntax, I would slightly prefer \ifnot to \if_not or worse \unless. I would disaprove strongly of \unless because it looses the clear symmetry with a closing \fi.
--
Fabien.
Pavel had previously written this patch http://ftp.pgpi.org/pub/databases/postgresql/projects/pgFoundry/epsql/epsql/epsql-8.5-develop/psql-enhanced-macros.diff which I leave here as a point of reference.
Obviously, there's a lot more in there than we'd need, and it's back quite a few versions.
I agree that the boolean tests available should be *very* simple, and all of the weight of complex calculation should be put in SQL, like we do with \gset
I propose those be:
\if STRING : true if STRING evaluates to true via ParseVariableBool, empty means false\ifnot STRING: true if STRING evaluates to false via ParseVariableBool, empty means false\ifdef psql_var: true if psql_var is defined\ifndef psql_var: true if psql_var is not defined, helpful for when --set var=val was omitted and should be defaultedA full compliment of \elseif \elseifnot \elseifdef \elseifndef, each matching the corresponding \if* above. I'm ok with these being optional in the first revision.\else - technically we could leave this out as well\endif
Then seems like we need an if-state-stack to handle nesting. At any given point, the mode could be:
1. Nonepsql currently isn't in an if-then structure, no non-conditional statements are skipped. any conditional commands other than \if* are an error.For that matter, the script can always add a new level to the stack with an \if* command.2. If-Thenpsql is currently executing statements in an \if* branch that evaluated to true. valid conditional commands are: \if*, \else* \endif3. If-Skippsql is currently in a block that evaluated to false, and will still parse commands for psql-correctness, but will skip them until it encounters an \endif or \else*4. Else-ThenJust like If-Then, but encountering an \else* command would be an error.5. Else-SkipJust like If-Skip, but encountering an \else* command would be an error.
The only data structure we'd need is the stack of enums listed above. Any commands would check against the stack-state before executing, but would otherwise be parsed and checked as they are now. The new \commands would manipulate that stack.
Does that seem work-able?
Hello Corey, > I agree that the boolean tests available should be *very* simple, and all > of the weight of complex calculation should be put in SQL, like we do with > \gset Yes. > I propose those be: > > \if STRING : true if STRING evaluates to true via ParseVariableBool, empty > means false Yep. > \ifnot STRING: true if STRING evaluates to false via ParseVariableBool, > empty means false I'd like other opinions about having unique \if and an not operator, vs multiplying \if* and possibly \elif* keywords. > \ifdef psql_var: true if psql_var is defined > \ifndef psql_var: true if psql_var is not defined, helpful for when --set > var=val was omitted and should be defaulted Hmmm. Would you have an example use case that could not be done simply with the previous ifs? cpp did that end ended up with a single if in the end. > A full compliment of \elseif \elseifnot \elseifdef \elseifndef, each > matching the corresponding \if* above. I'm ok with these being optional in > the first revision. > \else - technically we could leave this out as well > \endif For consistency, the possible sources of inspiration for a syntax with an explicit end marker are: - PL/pgSQL: if / then / elsif / else / endif - cpp: if / elif / else / endif - sh: if / then / elif / else / fi Now "then" is useless in a line oriented syntax, for which the closest example above is cpp, which does not have it. I think that we should stick to one of these. I like the shell syntax (without then), but given the line orientation maybe it would make sense to use the cpp version, close to what you are proposing. I cannot remember a language with elseif* variants, and I find them quite ugly, so from an aethetical point of view I would prefer to avoid that... On the other hand having an "else if" capability makes sense (eg do something slightly different for various versions of pg), so that would suggest to stick to a simpler "if" without variants, if possible. > Then seems like we need an if-state-stack to handle nesting. [...] Yes, a stack or recursion is needed for implementing nesting. > States: None, If-Then, If-Skip, Else-Then, Else-Skip. With an "else if" construct you probably need some more states: You have to know whether you already executed a block in which case a subsequent condition is ignored, so there is a state "skip all to end" needed. > Does that seem work-able? Sure. I think the priority is to try to agree about a syntax, the implementation is a detail. -- Fabien.
Hmmm. Would you have an example use case that could not be done simply with the previous ifs? cpp did that end ended up with a single if in the end.
I think this is what you're asking for...
$ cat not_defined.sqlselect :'foo';$ psql postgres --no-psqlrc -f not_defined.sql --set foo=bar?column?----------bar(1 row)$ psql postgres --no-psqlrc -f not_defined.sqlpsql:not_defined.sql:3: ERROR: syntax error at or near ":"LINE 1: select :'foo';^
Now, if we instead added a way for psql to test whether or not a psql var was defined and set that boolean as ANOTHER variable, then we could avoid \ifdef and \ifndef.
For consistency, the possible sources of inspiration for a syntax with an explicit end marker are:
- PL/pgSQL: if / then / elsif / else / endif
- cpp: if / elif / else / endif
- sh: if / then / elif / else / fi
Now "then" is useless in a line oriented syntax, for which the closest example above is cpp, which does not have it. I think that we should stick to one of these.
I like the shell syntax (without then), but given the line orientation maybe it would make sense to use the cpp version, close to what you are proposing.
I think we should use pl/pgsql as our inspiration, though there's no need for the "then" because psql commands end the line....which makes it identical to the C++ version.
But if we can get this thing done, I really don't care which we use.
I cannot remember a language with elseif* variants, and I find them quite ugly, so from an aethetical point of view I would prefer to avoid that... On the other hand having an "else if" capability makes sense (eg do something slightly different for various versions of pg), so that would suggest to stick to a simpler "if" without variants, if possible.
We need to keep things easy to parse. Earlier someone said no psql command should ever have more than 2 parameters, and generally only one. Increasing the number of commands allows us to avoid multi-parameter commands. So it's a trade-off, we have more, simpler commands, or fewer, more complicated ones.
\if [not] [defined] [<string>]\elsif [not] [defined] [<string>]
is problematic if string is ever "not" or "defined". If someone can show me a way around that, I'm game.
Then seems like we need an if-state-stack to handle nesting. [...]
Yes, a stack or recursion is needed for implementing nesting.States: None, If-Then, If-Skip, Else-Then, Else-Skip.
With an "else if" construct you probably need some more states: You have to know whether you already executed a block in which case a subsequent condition is ignored, so there is a state "skip all to end" needed.
Right, we'd have to check every level of the stack for a skip-state, not a big deal.
On 11/29/2016 03:07 PM, Fabien COELHO wrote: > > > I cannot remember a language with elseif* variants, and I find them > quite ugly, so from an aethetical point of view I would prefer to > avoid that... On the other hand having an "else if" capability makes > sense (eg do something slightly different for various versions of pg), > so that would suggest to stick to a simpler "if" without variants, if > possible. > > FTR I *strongly* disagree with this. (And if you can't remember a language that comes with them then you need to get out more. The Bourne shell, where it's spelled "elif", and Ada are two obvious examples.) cheers andrew
On 11/29/16 5:08 PM, Andrew Dunstan wrote: > > On 11/29/2016 03:07 PM, Fabien COELHO wrote: >> >> >> I cannot remember a language with elseif* variants, and I find them >> quite ugly, so from an aethetical point of view I would prefer to >> avoid that... On the other hand having an "else if" capability makes >> sense (eg do something slightly different for various versions of pg), >> so that would suggest to stick to a simpler "if" without variants, if >> possible. > > FTR I *strongly* disagree with this. (And if you can't remember a > language that comes with them then you need to get out more. The Bourne > shell, where it's spelled "elif", and Ada are two obvious examples.) Not to mention PL/pgSQL and Perl. -- -David david@pgmasters.net
On 11/29/2016 05:25 PM, David Steele wrote: > On 11/29/16 5:08 PM, Andrew Dunstan wrote: >> On 11/29/2016 03:07 PM, Fabien COELHO wrote: >>> >>> I cannot remember a language with elseif* variants, and I find them >>> quite ugly, so from an aethetical point of view I would prefer to >>> avoid that... On the other hand having an "else if" capability makes >>> sense (eg do something slightly different for various versions of pg), >>> so that would suggest to stick to a simpler "if" without variants, if >>> possible. >> FTR I *strongly* disagree with this. (And if you can't remember a >> language that comes with them then you need to get out more. The Bourne >> shell, where it's spelled "elif", and Ada are two obvious examples.) > Not to mention PL/pgSQL and Perl. > Indeed. cheers andrew
On Tue, Nov 29, 2016 at 01:10:06PM +0100, Fabien COELHO wrote: > If there is a negative condition syntax, I would slightly prefer \ifnot to > \if_not or worse \unless. I would disaprove strongly of \unless because it > looses the clear symmetry with a closing \fi. I take it \sselnu is right out. Best, David. -- David Fetter <david(at)fetter(dot)org> http://fetter.org/ Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com Remember to vote! Consider donating to Postgres: http://www.postgresql.org/about/donate
David Fetter <david@fetter.org> writes:
> On Tue, Nov 29, 2016 at 01:10:06PM +0100, Fabien COELHO wrote:
>> If there is a negative condition syntax, I would slightly prefer \ifnot to
>> \if_not or worse \unless. I would disaprove strongly of \unless because it
>> looses the clear symmetry with a closing \fi.
> I take it \sselnu is right out.
[ splorf... ] But really, \fi is something that could only be loved
by a certain academic breed of hackers. I'd go for \endif, probably.
That still doesn't relate well with \unless, so +1 for \if, \ifnot,
\else, and \endif.
I'm not really convinced that we need an \elseif at this point.
It could be added later if people find it compelling, but I'm
having a hard time believing that it's essential.
regards, tom lane
Hello Tom, > But really, \fi is something that could only be loved by a certain > academic breed of hackers. Ah ah:-) I'll take that as a compliment:-) > I'd go for \endif, probably. That still doesn't relate well with > \unless, so +1 for \if, \ifnot, \else, and \endif. Ok, that is a clear opinion. > I'm not really convinced that we need an \elseif at this point. > It could be added later if people find it compelling, but I'm > having a hard time believing that it's essential. My experience with cpp is that #elif is quite useful in some cases, typically with alternate implementations depending on dependences, to avoid this: #ifdef HAVE_XXX ... #else #ifdef HAVE_YYY ... #else #ifdef HAVE ZZZ ... #else #error "oops!" #endif // HAVE_ZZZ #endif// HAVE_YYY #endif // HAVE_XXX I think that some realistic use cases examples for psql would be great... If the else-if is a common pattern then it should be at least in the design, even if not implemented right away. Also ISTM that having else-if is not too compatible with multiplying \if* because it leads to pretty ugly \el*if* variants which are quite hard to parse and understand, so there is an impact on the design. -- Fabien.
Hello Andrew, >> I cannot remember a language with elseif* variants, and I find them quite >> ugly, so from an aethetical point of view I would prefer to avoid that... >> On the other hand having an "else if" capability makes sense (eg do >> something slightly different for various versions of pg), so that would >> suggest to stick to a simpler "if" without variants, if possible. > > FTR I *strongly* disagree with this. (And if you can't remember a language > that comes with them then you need to get out more. The Bourne shell, where > it's spelled "elif", and Ada are two obvious examples.) There may be a misunderstanding somewhere. I'm rather in favor of having "elif/elsif/elseif/..." constructs, especially if they can be useful in realistic examples, which is not clear yet for psql scripts. I'm arguing against "if/elif" *variants* in the sense of various conditional semantics: e.g. in cpp you have several "if"s (ifdef ifndef if), but you do not have all the corresponding "elif"s (elifdef elifndef...), there is only one "elif". In cpp "ifdef"/"ifndef" were obsoleted by "if" with minimal expression support (#if !defined(HAS_SOMETHING) ...) and only this "if" has its "elif". -- Fabien.
On 11/30/2016 03:47 AM, Fabien COELHO wrote:
>
> Hello Andrew,
>
>>> I cannot remember a language with elseif* variants, and I find them
>>> quite ugly, so from an aethetical point of view I would prefer to
>>> avoid that... On the other hand having an "else if" capability makes
>>> sense (eg do something slightly different for various versions of
>>> pg), so that would suggest to stick to a simpler "if" without
>>> variants, if possible.
>>
>> FTR I *strongly* disagree with this. (And if you can't remember a
>> language that comes with them then you need to get out more. The
>> Bourne shell, where it's spelled "elif", and Ada are two obvious
>> examples.)
>
> There may be a misunderstanding somewhere.
>
> I'm rather in favor of having "elif/elsif/elseif/..." constructs,
> especially if they can be useful in realistic examples, which is not
> clear yet for psql scripts.
>
> I'm arguing against "if/elif" *variants* in the sense of various
> conditional semantics: e.g. in cpp you have several "if"s (ifdef
> ifndef if), but you do not have all the corresponding "elif"s (elifdef
> elifndef...), there is only one "elif". In cpp "ifdef"/"ifndef" were
> obsoleted by "if" with minimal expression support (#if
> !defined(HAS_SOMETHING) ...) and only this "if" has its "elif".
>
Oh, I see. Sorry for the misunderstanding.
I agree about generally sticking with one pattern, but I wouldn't want
to exclude shorthand pieces like
\quit_if cond
which could be more convenient than
\if cond \quit \endif
c.f. perl's "foo if bar;" as shorthand for "if (bar) { foo; }"
Still, that might be a refinement to add later on.
cheers
andrew
On Tue, Nov 29, 2016 at 4:43 AM, Pavel Stehule <pavel.stehule@gmail.com> wrote: > I prefer the commands instead symbols - the parsing and processing symbols > should be more complex than it is now. A psql parser is very simple - and > any complex syntax enforces lot of code. > > \if_not Given the precedent in pgbench (cf. 878fdcb843e087cc1cdeadc987d6ef55202ddd04), I think it requires an amazing level of optimism to suppose we won't eventually end up with a full-blown expression language here. I would suggest designing one from the beginning and getting it over with. Even if you manage to hold the line and exclude it from whatever gets committed initially, somebody's going to propose it 2 years from now. And again 4 years from now. And again 2 years after that. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
Robert Haas <robertmhaas@gmail.com> writes:
> Given the precedent in pgbench (cf.
> 878fdcb843e087cc1cdeadc987d6ef55202ddd04), I think it requires an
> amazing level of optimism to suppose we won't eventually end up with a
> full-blown expression language here. I would suggest designing one
> from the beginning and getting it over with. Even if you manage to
> hold the line and exclude it from whatever gets committed initially,
> somebody's going to propose it 2 years from now. And again 4 years
> from now. And again 2 years after that.
The other problem with not thinking about that general case is that
people will keep on proposing little bitty features that nibble at
the problem but may or may not be compatible with a general solution.
To the extent that such patches get accepted, we'll be forced into
either backwards-compatibility breakage or sub-optimal solutions when
we do get to the point of wanting a general answer. I'd much rather
start with a generalized design and then implement it piece by piece.
(This is more or less the same point as my nearby stand against localized
hacking of backslash parsing rules.)
regards, tom lane
<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px#ccc solid;padding-left:1ex"><span class=""><br /></span>The other problem with not thinking about thatgeneral case is that<br /> people will keep on proposing little bitty features that nibble at<br /> the problem but mayor may not be compatible with a general solution.<br /> To the extent that such patches get accepted, we'll be forcedinto<br /> either backwards-compatibility breakage or sub-optimal solutions when<br /> we do get to the point of wantinga general answer. I'd much rather<br /> start with a generalized design and then implement it piece by piece.<br/><br /> (This is more or less the same point as my nearby stand against localized<br /> hacking of backslash parsingrules.)<br /><br /> regards, tom lane<br /></blockquote></div><br /></div><div class="gmail_extra"><br/></div><div class="gmail_extra">In order for me to understand how high the bar has been set, canyou (Robert/Tom mostly, but I welcome any responses) explain what you mean by "full-blown expression language"? What constructsmust it include, etc?</div><div class="gmail_extra"><br /></div><div class="gmail_extra"><br /></div><div class="gmail_extra"><br/></div><div class="gmail_extra"><br /></div><div class="gmail_extra"><br /></div></div>
Corey Huinker <corey.huinker@gmail.com> writes:
> In order for me to understand how high the bar has been set, can you
> (Robert/Tom mostly, but I welcome any responses) explain what you mean by
> "full-blown expression language"? What constructs must it include, etc?
My guess is that something comparable to where pgbench is would be a
reasonable target --- not least because I think we should strive to
reduce unnecessary differences between psql and pgbench metalanguages.
I'm not sure that I'm ready to propose that they should share the same
expression engine, but perhaps it's not a totally wacky idea.
regards, tom lane
On Fri, Dec 2, 2016 at 11:12 AM, Corey Huinker <corey.huinker@gmail.com> wrote: > In order for me to understand how high the bar has been set, can you > (Robert/Tom mostly, but I welcome any responses) explain what you mean by > "full-blown expression language"? What constructs must it include, etc? I mostly mean it should be based on flex and bison, not "this is so simple I can just hand-roll it". I don't think it has much chance of staying that simple. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
On 12/2/16 9:24 AM, Robert Haas wrote: > On Fri, Dec 2, 2016 at 11:12 AM, Corey Huinker <corey.huinker@gmail.com> wrote: >> In order for me to understand how high the bar has been set, can you >> (Robert/Tom mostly, but I welcome any responses) explain what you mean by >> "full-blown expression language"? What constructs must it include, etc? > > I mostly mean it should be based on flex and bison, not "this is so > simple I can just hand-roll it". I don't think it has much chance of > staying that simple. It might be worth looking at other simplistic languages for guidance. bash comes to mind, though there's GPL issues there. csh/tcsh don't have those problems, and perhaps some of their grammar could be stolen. I find it interesting that this is kind of the opposite problem that most pl's face: how to fit the database access paradigm into the language with the minimal amount of extra effort for users. http://cvsweb.netbsd.org/bsdweb.cgi/src/bin/csh/?only_with_tag=MAIN https://github.com/tcsh-org/tcsh -- Jim Nasby, Data Architect, Blue Treble Consulting, Austin TX Experts in Analytics, Data Architecture and PostgreSQL Data in Trouble? Get it in Treble! http://BlueTreble.com 855-TREBLE2 (855-873-2532)
Hello,
> My guess is that something comparable to where pgbench is would be a
> reasonable target --- not least because I think we should strive to
> reduce unnecessary differences between psql and pgbench metalanguages.
>
> I'm not sure that I'm ready to propose that they should share the same
> expression engine, but perhaps it's not a totally wacky idea.
I'm trying to summarize a proposal for a conditional structure:
- existing psql ":"-variables can be used, allowing anything from SQL (eg querying about available objects, features,
extensions, current settings...)
- target psql conditional syntax could be:
\if <expression> ... \elif <...> ... \else ... \endif
- possible incremental implemention steps on this path:
(1) minimal condition and expression, compatible with a possible future full-blown expression syntax
\if :variable \if not :variable -- maybe \if ! :variable ... \endif
(2) add "\else"
(3) add "\elif ..." (or maybe "\elsif ..."?)
(4) add greater but limited expressions, compatible with a full blown expression syntax (eg \if :var/const
<comparison-operator>:var/const)
(5) add full-blown <expression> support for \if, which suggest that it would also be available for \set
Does this looks okay, or does it need to be amended?
A few comments:
Given the experience with pgbench and the psql context, I do not think
that it would really need to go beyond step 2 above, but I agree that I
may be wrong and it is best to be prepared for that from the start. Given
the complexity and effort involved with (5), it seems wise to wait for a
clearer motivation with actual use-cases before going that far.
In the benchmarking context, the point is to test performance for a
client-server scenario, in which case the client has to do some things,
thus needs miminal computation capabilities which were available early in
pgbench (\setrandom, \set with one arithmetic operation...) because they
were necessary. Probably \if ... would make sense in pgbench, so I'll
think about it.
In psql interactive context, conditions and expressions do not make sense
as the user typing the command knows what they want, thus will do
evaluations in their head to avoid typing stuff...
In psql scripting context, conditions are likely to be about what to do
with the database, and what I understand of the use-case which started
this discussion is that it is about versions, settings, available objects,
typically "if this is already installed, skip this part" or "if version
beyond YYY, cannot install because of missing features" when installing
and initializing an application. For this purpose, ISTM that the query is
necessarily performed server-side, thus the actual need for a full-blown
client-side expression is limited or void, although as already said being
prepared is a good thing.
--
Fabien.
2016-12-03 8:16 GMT+01:00 Fabien COELHO <coelho@cri.ensmp.fr>:
Hello,My guess is that something comparable to where pgbench is would be a
reasonable target --- not least because I think we should strive to
reduce unnecessary differences between psql and pgbench metalanguages.
I'm not sure that I'm ready to propose that they should share the same
expression engine, but perhaps it's not a totally wacky idea.
I'm trying to summarize a proposal for a conditional structure:
- existing psql ":"-variables can be used, allowing anything from SQL
(eg querying about available objects, features, extensions,
current settings...)
- target psql conditional syntax could be:
\if <expression>
...
\elif <...>
...
\else
...
\endif
- possible incremental implemention steps on this path:
(1) minimal condition and expression, compatible with
a possible future full-blown expression syntax
\if :variable
\if not :variable -- maybe \if ! :variable
...
\endif
(2) add "\else"
(3) add "\elif ..." (or maybe "\elsif ..."?)
(4) add greater but limited expressions, compatible with a full blown
expression syntax (eg \if :var/const <comparison-operator> :var/const)
(5) add full-blown <expression> support for \if, which suggest that
it would also be available for \set
Does this looks okay, or does it need to be amended?
A few comments:
Given the experience with pgbench and the psql context, I do not think that it would really need to go beyond step 2 above, but I agree that I may be wrong and it is best to be prepared for that from the start. Given the complexity and effort involved with (5), it seems wise to wait for a clearer motivation with actual use-cases before going that far.
In the benchmarking context, the point is to test performance for a client-server scenario, in which case the client has to do some things, thus needs miminal computation capabilities which were available early in pgbench (\setrandom, \set with one arithmetic operation...) because they were necessary. Probably \if ... would make sense in pgbench, so I'll think about it.
In psql interactive context, conditions and expressions do not make sense as the user typing the command knows what they want, thus will do evaluations in their head to avoid typing stuff...
In psql scripting context, conditions are likely to be about what to do with the database, and what I understand of the use-case which started this discussion is that it is about versions, settings, available objects, typically "if this is already installed, skip this part" or "if version beyond YYY, cannot install because of missing features" when installing and initializing an application. For this purpose, ISTM that the query is necessarily performed server-side, thus the actual need for a full-blown client-side expression is limited or void, although as already said being prepared is a good thing.
Some possibilities from pgbench can have sense in psql too - generating some random numbers from a range. In the end we use one parser for psql and for pgbench.
I agree, so step 2 should be enough, and I accept so there is opened door for any future enhancing.
We can implement some client side boolean functions (similar to pgbench functions that can cover often tasks: version_less, version_greather, user_exists, tables_exists, index_exists, variable_exists, schema_exists,
Regards
Pavel
--
Fabien.
Hello Pavel, > Some possibilities from pgbench can have sense in psql too - generating > some random numbers from a range. Could you expand on the use case where this would be useful? > In the end we use one parser for psql and for pgbench. Note that "master" lexer is already shared, thanks to Tom, so as to detect consistently where a query ends. > I agree, so step 2 should be enough, and I accept so there is opened door > for any future enhancing. Good, because that was the idea:-) > We can implement some client side boolean functions (similar to pgbench > functions that can cover often tasks: version_less, version_greather, > user_exists, tables_exists, index_exists, variable_exists, schema_exists, Yes, that is a possibility, but this can already be queried into a :-variable, so it is less indispensable. -- Fabien.
2016-12-04 17:35 GMT+01:00 Fabien COELHO <coelho@cri.ensmp.fr>:
Hello Pavel,Some possibilities from pgbench can have sense in psql too - generating
some random numbers from a range.
Could you expand on the use case where this would be useful?
writing test cases
In the end we use one parser for psql and for pgbench.
Note that "master" lexer is already shared, thanks to Tom, so as to detect consistently where a query ends.I agree, so step 2 should be enough, and I accept so there is opened door
for any future enhancing.
Good, because that was the idea:-)We can implement some client side boolean functions (similar to pgbench
functions that can cover often tasks: version_less, version_greather,
user_exists, tables_exists, index_exists, variable_exists, schema_exists,
Yes, that is a possibility, but this can already be queried into a :-variable, so it is less indispensable.
can you show some examples, please?
Regards
Pavel
--
Fabien.
>> Yes, that is a possibility, but this can already be queried into a >> :-variable, so it is less indispensable. > > can you show some examples, please? SELECT COUNT(*) AS has_unit_extension FROM pg_extension WHERE extname='unit' \gset \echo :has_unit_extension 1 So that \if ! :hash_unit_extension CREATE TABLE foo(id SERIAL, stuff UNIT); \else \echo "unit extension is not loaded" \quit \fi Ok, for this example one may try: CREATE EXTENSION IF NOT EXISTS unit; Or use the "ON_ERROR_STOP" setting, but that is the idea, SQL can be used to test anything server-side. -- Fabien.
2016-12-04 20:55 GMT+01:00 Fabien COELHO <coelho@cri.ensmp.fr>:
Yes, that is a possibility, but this can already be queried into a :-variable, so it is less indispensable.
can you show some examples, please?
SELECT COUNT(*) AS has_unit_extension
FROM pg_extension WHERE extname='unit' \gset
\echo :has_unit_extension
1
So that
\if ! :hash_unit_extension
CREATE TABLE foo(id SERIAL, stuff UNIT);
\else
\echo "unit extension is not loaded"
\quit
\fi
Ok, for this example one may try:
CREATE EXTENSION IF NOT EXISTS unit;
Or use the "ON_ERROR_STOP" setting, but that is the idea, SQL can be used to test anything server-side.
understand
I am thinking so first step can be simply and much more friendly replaced by specialized function:
\if has_extension(...)
the functions are supported by pgbench already, so we can take code from there.
I don't think so psql script language should be too rich - it should be DSL and often use cases should be supported with friendly syntax.
The set of functions can be small in first stage - we can support only one function.
Regards
Pavel
--
Fabien.
On Sat, Dec 3, 2016 at 2:16 AM, Fabien COELHO <coelho@cri.ensmp.fr> wrote: >> My guess is that something comparable to where pgbench is would be a >> reasonable target --- not least because I think we should strive to >> reduce unnecessary differences between psql and pgbench metalanguages. >> >> I'm not sure that I'm ready to propose that they should share the same >> expression engine, but perhaps it's not a totally wacky idea. > > > I'm trying to summarize a proposal for a conditional structure: > > - existing psql ":"-variables can be used, allowing anything from SQL > (eg querying about available objects, features, extensions, > current settings...) > > - target psql conditional syntax could be: > > \if <expression> > ... > \elif <...> > ... > \else > ... > \endif > > - possible incremental implemention steps on this path: > > (1) minimal condition and expression, compatible with > a possible future full-blown expression syntax > > \if :variable > \if not :variable -- maybe \if ! :variable > ... > \endif > > (2) add "\else" > > (3) add "\elif ..." (or maybe "\elsif ..."?) > > (4) add greater but limited expressions, compatible with a full blown > expression syntax (eg \if :var/const <comparison-operator> :var/const) > > (5) add full-blown <expression> support for \if, which suggest that > it would also be available for \set > > > Does this looks okay, or does it need to be amended? > > A few comments: > > Given the experience with pgbench and the psql context, I do not think that > it would really need to go beyond step 2 above, but I agree that I may be > wrong and it is best to be prepared for that from the start. Given the > complexity and effort involved with (5), it seems wise to wait for a clearer > motivation with actual use-cases before going that far. Well, my vote would be to go all the way to #5 in one commit. Stopping short of that doesn't seem to me to save enough work to make much sense. I don't think we're talking about anything all that complex, and it will make future improvements a lot simpler. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
Hello Robert, >> Given the experience with pgbench and the psql context, I do not think that >> it would really need to go beyond step 2 above, but I agree that I may be >> wrong and it is best to be prepared for that from the start. Given the >> complexity and effort involved with (5), it seems wise to wait for a clearer >> motivation with actual use-cases before going that far. > > Well, my vote would be to go all the way to #5 in one commit. > Stopping short of that doesn't seem to me to save enough work to make > much sense. I don't think we're talking about anything all that > complex, and it will make future improvements a lot simpler. First, my experience as a basic patch submitter is that any patch which does more than one thing at a time, even somehow closely related changes, is asked to be split into distinct sub-patches, and is harder to get through. Second, requiring more advanced features is a recipee for getting nothing in the end, because even if not "that complex" it requires significant more time to develop. The first step I outlined is enough to handle the submitted use case and is compatible with grand plans which would change significantly psql, so seems a reasonnable intermediate target. Your experience as an seasoned core developer and a committer is probably different:-) -- Fabien.
On Tue, Dec 6, 2016 at 1:28 AM, Fabien COELHO <coelho@cri.ensmp.fr> wrote: > First, my experience as a basic patch submitter is that any patch which does > more than one thing at a time, even somehow closely related changes, is > asked to be split into distinct sub-patches, and is harder to get through. > > Second, requiring more advanced features is a recipee for getting nothing in > the end, because even if not "that complex" it requires significant more > time to develop. The first step I outlined is enough to handle the submitted > use case and is compatible with grand plans which would change significantly > psql, so seems a reasonnable intermediate target. > > Your experience as an seasoned core developer and a committer is probably > different:-) Well, everybody can have their own opinion on what is reasonable. There are times I argue for making a patch smaller (frequent) and times when I argue for making it bigger (rare). We had pretty much this exact same argument about the pgbench lexer and parser and in the event I coded something up in less than a day. This argument feels like a rerun of that one. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
Hello Robert, >> Your experience as an seasoned core developer and a committer is >> probably different:-) > > Well, everybody can have their own opinion on what is reasonable. Sure. > There are times I argue for making a patch smaller (frequent) and > times when I argue for making it bigger (rare). Yep. > We had pretty much this exact same argument about the pgbench lexer and > parser and in the event I coded something up in less than a day. This > argument feels like a rerun of that one. There are some differences: pgbench already had one operator at a time expressions, while psql has survived till today with none, which suggest a less pressing need. Moreover the features are partly orthogonal and would touch psql significantly in different although probably overlapping areas: - expressions is rather about \set, even if reused with \ifas well - condition is about \if ... \endif and ignoring some input lines The current expression evaluation in pgbench is about 1000 lines for scanning, parsing & evaluating, and does not yet support boolean expressions, although a patch for that has been in the queue for some time. I foresee that someone will suggest/require/demand... that the expression code be shared between pgbench and psql, which is another argument for dissociating these two features (expression and conditional in psql) from the start. -- Fabien.
On Tue, Dec 6, 2016 at 8:45 AM, Fabien COELHO <coelho@cri.ensmp.fr> wrote: > There are some differences: pgbench already had one operator at a time > expressions, while psql has survived till today with none, which suggest a > less pressing need. I don't really think so. People have been wanting expressions in psql since I got involved in the project. > Moreover the features are partly orthogonal and would touch psql > significantly in different although probably overlapping areas: > - expressions is rather about \set, even if reused with \if as well > - condition is about \if ... \endif and ignoring some input lines I don't think that difference is very relevant, really. > The current expression evaluation in pgbench is about 1000 lines for > scanning, parsing & evaluating, and does not yet support boolean > expressions, although a patch for that has been in the queue for some time. > I foresee that someone will suggest/require/demand... that the expression > code be shared between pgbench and psql, which is another argument for > dissociating these two features (expression and conditional in psql) from > the start. That seems like an argument the other way, from here. I'm not sure that Tom is right in wanting so much to be shared between psql and pgbench. They're different tools with different purposes, and I'm not sure sharing between them makes much sense. But if it does make sense to share, then that's another reason for not designing something ad-hoc for psql: integrating it later will be more work in total. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
> But if it does make sense to share, then that's another reason for not > designing something ad-hoc for psql: integrating it later will be more > work in total. Yes, but not much: evaluating "[!] :var" special syntax would be dropped, but I do not think that it is the main issue with these features. I understand that your conclusion is to require psql expressions + conditions as one (large) patch. Given my personnal experience with the patch process, my conclusion is that I will very probably not do it. Just adding logical expressions to pgbench, a minor feature for a minor client, has already been spread over 3 CF, and this is a part of what is required for the condition & expression in psql. Hopefully someone else will do it, and I'll do some reviewing then. -- Fabien.
On Mon, Dec 5, 2016 at 12:32 PM, Robert Haas <robertmhaas@gmail.com> wrote: >> - possible incremental implemention steps on this path: >> >> (1) minimal condition and expression, compatible with >> a possible future full-blown expression syntax >> >> \if :variable >> \if not :variable -- maybe \if ! :variable >> ... >> \endif >> >> (2) add "\else" >> >> (3) add "\elif ..." (or maybe "\elsif ..."?) >> >> (4) add greater but limited expressions, compatible with a full blown >> expression syntax (eg \if :var/const <comparison-operator> :var/const) >> >> (5) add full-blown <expression> support for \if, which suggest that >> it would also be available for \set >> >> >> Does this looks okay, or does it need to be amended? >> >> A few comments: >> >> Given the experience with pgbench and the psql context, I do not think that >> it would really need to go beyond step 2 above, but I agree that I may be >> wrong and it is best to be prepared for that from the start. Given the >> complexity and effort involved with (5), it seems wise to wait for a clearer >> motivation with actual use-cases before going that far. > > Well, my vote would be to go all the way to #5 in one commit. > Stopping short of that doesn't seem to me to save enough work to make > much sense. I don't think we're talking about anything all that > complex, and it will make future improvements a lot simpler. After having thought about this a little bit further and reread this a bit more carefully, I would like to revise my position. Really, what I don't want to end up with is a hand-coded expression syntax that is very limited which then has to be replaced with a full-blown lexer and parser. That is, I do not want to ever be at step "4" of this proposal. So I think it would be reasonable for somebody to implement \if, \elseif, \endif first, with the argument having to be, precisely, a single variable and nothing else (not even a negator). Then a future patch could allow an expression there instead of a variable. I don't think that would be any harder to review than going all the way to #5 in one shot, and actually it might be simpler. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
On Mon, Dec 5, 2016 at 12:32 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>> - possible incremental implemention steps on this path:
>>
>> (1) minimal condition and expression, compatible with
>> a possible future full-blown expression syntax
>>
>> \if :variable
>> \if not :variable -- maybe \if ! :variable
We don't presently have a unary boolean operator named "!" so adding this variant would create an inconsistency
So I think it would be reasonable for somebody to implement \if,
\elseif, \endif first, with the argument having to be, precisely, a
single variable and nothing else (not even a negator). Then a future
patch could allow an expression there instead of a variable. I don't
think that would be any harder to review than going all the way to #5
in one shot, and actually it might be simpler.
I worry about the case of disallowing negation in #1 and then not getting to #5 (in the same version) where the expression "not(var)" becomes possible.
If the expected committed patch set includes #5 then this becomes a matter for reviewer convenience so never mind. But if its at all possible for #5 to be punted down the road incorporating the eventual "not var" and "not(var)" syntax into #1 as a kind of shim would seem desirable.
David J.
2016-12-16 18:21 GMT+01:00 David G. Johnston <david.g.johnston@gmail.com>:
On Mon, Dec 5, 2016 at 12:32 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>> - possible incremental implemention steps on this path:
>>
>> (1) minimal condition and expression, compatible with
>> a possible future full-blown expression syntax
>>
>> \if :variable
>> \if not :variable -- maybe \if ! :variableWe don't presently have a unary boolean operator named "!" so adding this variant would create an inconsistency
If we allow some complex expressions there, then it should be a SQL expressions evaluated on server side.
There are two variants - 1. simple client side expression - can be functional only, 2. complex server side expression.
So I think it would be reasonable for somebody to implement \if,
\elseif, \endif first, with the argument having to be, precisely, a
single variable and nothing else (not even a negator). Then a future
patch could allow an expression there instead of a variable. I don't
think that would be any harder to review than going all the way to #5
in one shot, and actually it might be simpler.I worry about the case of disallowing negation in #1 and then not getting to #5 (in the same version) where the expression "not(var)" becomes possible.If the expected committed patch set includes #5 then this becomes a matter for reviewer convenience so never mind. But if its at all possible for #5 to be punted down the road incorporating the eventual "not var" and "not(var)" syntax into #1 as a kind of shim would seem desirable.
why do you need special operator for negation? there is only one use case. It can be solved by \if_not
David J.
2016-12-16 18:21 GMT+01:00 David G. Johnston <david.g.johnston@gmail.com>:If the expected committed patch set includes #5 then this becomes a matter for reviewer convenience so never mind. But if its at all possible for #5 to be punted down the road incorporating the eventual "not var" and "not(var)" syntax into #1 as a kind of shim would seem desirable.why do you need special operator for negation? there is only one use case. It can be solved by \if_not
Not following the thread that closely and the section Robert quoted didn't include "\if_not" as a syntax option. I figured the idea was to limit the number of backslash commands and leave the power in the expression evaluation.
David J.
Robert Haas <robertmhaas@gmail.com> writes:
> So I think it would be reasonable for somebody to implement \if,
> \elseif, \endif first, with the argument having to be, precisely, a
> single variable and nothing else (not even a negator). Then a future
> patch could allow an expression there instead of a variable. I don't
> think that would be any harder to review than going all the way to #5
> in one shot, and actually it might be simpler.
This seems like a reasonable implementation plan to me, not least because
it tackles the hard part first. There's no doubt that we can build an
expression evaluator, but I'm not entirely sure how we're going to wedge
conditional eval or loops into psql's command reader.
(Or in other words, let's see \while ... \endloop in the minimal proposal
as well, or at least a sketch of how to get there.)
regards, tom lane
2016-12-16 18:33 GMT+01:00 David G. Johnston <david.g.johnston@gmail.com>:
2016-12-16 18:21 GMT+01:00 David G. Johnston <david.g.johnston@gmail.com>:If the expected committed patch set includes #5 then this becomes a matter for reviewer convenience so never mind. But if its at all possible for #5 to be punted down the road incorporating the eventual "not var" and "not(var)" syntax into #1 as a kind of shim would seem desirable.why do you need special operator for negation? there is only one use case. It can be solved by \if_notNot following the thread that closely and the section Robert quoted didn't include "\if_not" as a syntax option. I figured the idea was to limit the number of backslash commands and leave the power in the expression evaluation.
without a expression you can store a negation to variable
I can imagine simple functional only expressions evaluated on client side.
\if not(table_exists('table_name'))
full expressions are not easy implemented without bigger changes in psql parser design - and I don't see any reason why do some too complex there. I would not to replace bash, perl, python or lua.
David J.
On Fri, Dec 16, 2016 at 12:28 PM, Pavel Stehule <pavel.stehule@gmail.com> wrote: > why do you need special operator for negation? there is only one use case. > It can be solved by \if_not That's exactly the kind of thing I *don't* want to do. If you absolutely must have that and you can't wait until we get a full-blown expression evaluator, then just swap the \if side with the \else side and call it good. The whole point here is to avoid introducing weird hacks for lack of a full expression evaluator that will just become annoyances once we have one. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
2016-12-16 21:18 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:
On Fri, Dec 16, 2016 at 12:28 PM, Pavel Stehule <pavel.stehule@gmail.com> wrote:
> why do you need special operator for negation? there is only one use case.
> It can be solved by \if_not
That's exactly the kind of thing I *don't* want to do. If you
absolutely must have that and you can't wait until we get a full-blown
expression evaluator, then just swap the \if side with the \else side
and call it good. The whole point here is to avoid introducing weird
hacks for lack of a full expression evaluator that will just become
annoyances once we have one.
I don't need it. Because we don't expect expression there, then "not" or "if_not" is not necessary.
Regards
Pavel
Hello Tom, >> So I think it would be reasonable for somebody to implement \if, >> \elseif, \endif first, with the argument having to be, precisely, a >> single variable and nothing else (not even a negator). [...] > This seems like a reasonable implementation plan to me, not least because > it tackles the hard part first. There's no doubt that we can build an > expression evaluator, but I'm not entirely sure how we're going to wedge > conditional eval or loops into psql's command reader. > > (Or in other words, let's see \while ... \endloop in the minimal proposal > as well, or at least a sketch of how to get there.) My 0.02 €: I have not seen any use case for a loop... Does someone have something convincing? I could think of some use in benchmarking (aka in pgbench), but not psql... But I may lack imagination. If one realistic case is found, then from a syntactic point of view "\while expr ... \endwhile/loop/whatever" looks straightforward enough. However, the implementation issues are pretty different from "if" which can be managed pretty simply on the fly with a stack and a little automaton. A loop needs to store its body and evaluate it over and over, which means having processed the input up to the end of the loop before proceeding, including nesting and so... it is a much less interactive friendly construct. Note that although "cpp" has an if, but it does not have any loop. In my opinion psql should stay at that same simple level: ISTM that the typical psql-script requirement is to be able to test some things, eg for installing or upgrading the schema of an application, and for that variables, expressions server side and maybe client side, and conditions are mostly enough. A lot of "IF EXISTS" added to many commands recently are motivated to handle this kind of use-case at the command per command level, which is not necessarily the right place. A while loops turns a simple thing into a potential Turing-complete beast, without a strong incentive I think that it should be avoided. -- Fabien.
2016-12-17 16:26 GMT+01:00 Fabien COELHO <coelho@cri.ensmp.fr>:
+1
Hello Tom,So I think it would be reasonable for somebody to implement \if,
\elseif, \endif first, with the argument having to be, precisely, a
single variable and nothing else (not even a negator). [...]This seems like a reasonable implementation plan to me, not least because
it tackles the hard part first. There's no doubt that we can build an
expression evaluator, but I'm not entirely sure how we're going to wedge
conditional eval or loops into psql's command reader.
(Or in other words, let's see \while ... \endloop in the minimal proposal
as well, or at least a sketch of how to get there.)
My 0.02 €:
I have not seen any use case for a loop... Does someone have something convincing? I could think of some use in benchmarking (aka in pgbench), but not psql... But I may lack imagination.
If one realistic case is found, then from a syntactic point of view "\while expr ... \endwhile/loop/whatever" looks straightforward enough.
maybe iteration over cursor can be interesting - but now with with \gexec it is not important.
However, the implementation issues are pretty different from "if" which can be managed pretty simply on the fly with a stack and a little automaton. A loop needs to store its body and evaluate it over and over, which means having processed the input up to the end of the loop before proceeding, including nesting and so... it is a much less interactive friendly construct.
Note that although "cpp" has an if, but it does not have any loop.
In my opinion psql should stay at that same simple level: ISTM that the typical psql-script requirement is to be able to test some things, eg for installing or upgrading the schema of an application, and for that variables, expressions server side and maybe client side, and conditions are mostly enough. A lot of "IF EXISTS" added to many commands recently are motivated to handle this kind of use-case at the command per command level, which is not necessarily the right place.
A while loops turns a simple thing into a potential Turing-complete beast, without a strong incentive I think that it should be avoided.
+1
Pavel
--
Fabien.
On Fri, Dec 16, 2016 at 12:40 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Robert Haas <robertmhaas@gmail.com> writes: >> So I think it would be reasonable for somebody to implement \if, >> \elseif, \endif first, with the argument having to be, precisely, a >> single variable and nothing else (not even a negator). Then a future >> patch could allow an expression there instead of a variable. I don't >> think that would be any harder to review than going all the way to #5 >> in one shot, and actually it might be simpler. > > This seems like a reasonable implementation plan to me, not least because > it tackles the hard part first. There's no doubt that we can build an > expression evaluator, but I'm not entirely sure how we're going to wedge > conditional eval or loops into psql's command reader. > > (Or in other words, let's see \while ... \endloop in the minimal proposal > as well, or at least a sketch of how to get there.) It seems to me that we could implement \if ... \else ...\endif by having some kind of stack that indicates which constructs we're inside of and whether we're currently executing commands or discarding them. I think we want to avoid waiting until we see \endif to start running commands; it's better to execute or skip/discard each command as we see it. To implement \while, we'd also need to remember previous commands so that when we reach the end of the loop, we can rewind and put all of those commands back on the stack to be executed again, or perhaps to be skipped if the \while condition turns out now to be false. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
> (Or in other words, let's see \while ... \endloop in the minimal proposal
> as well, or at least a sketch of how to get there.)
It seems to me that we could implement \if ... \else ...\endif by
having some kind of stack that indicates which constructs we're inside
of and whether we're currently executing commands or discarding them.
I think we want to avoid waiting until we see \endif to start running
commands; it's better to execute or skip/discard each command as we
see it.
+1
To implement \while, we'd also need to remember previous commands so
that when we reach the end of the loop, we can rewind and put all of
those commands back on the stack to be executed again, or perhaps to
be skipped if the \while condition turns out now to be false.
This might be what you meant when you said "those commands back on the stack", but I think we'd have to remember not a list of commands, but the raw string of bytes from the start of the \while to the \endwhile (or equivalent), because any psql vars within that block could themselves be a non-parameter part of a command:
-- this is how I fake an 'exit 0' now:
\set work_needs_to_be_done t
select
select
case
when :'work_needs_to_be_done'::boolean then ''
else '\q'
end as cmd
\gset
:cmd
-- ridiculous example to illustrate complications in remembering past commands
select
case
when random() < 0.5 then '\ir my_script.sql'
when random() < 0.7 'select * from a_table; select count(*) from another_table;'
when random() < 0.7 'select * from a_table; select count(*) from another_table;'
else 'select null as foo;'
end as cmd
\gset
end as cmd
\gset
:cmd
And even then, things get complicated, because an \ir include which makes it this iteration might not make it the next, and the \endwhile might have been inside that include, or vice-versa, an included file starts a \while it doesn't finish.
So maybe what we store is a stack of buffers that are currently open (STDIN being captured as a buffer only when a \while starts, everything else being files), and additionally have a stack of positions where a \while started (buffer_id, position in buffer).
Additionally, we could assert that all \while-\endwhile pairs must happen in the same MainLoop (aka file), and mismatches are an error.
I'm happy to keep sketching out what control structures might look like and how to implement them. But I'm also happy to leave while/for loops out for now.
On Sat, Dec 17, 2016 at 3:39 PM, Corey Huinker <corey.huinker@gmail.com> wrote: >> To implement \while, we'd also need to remember previous commands so >> that when we reach the end of the loop, we can rewind and put all of >> those commands back on the stack to be executed again, or perhaps to >> be skipped if the \while condition turns out now to be false. > > This might be what you meant when you said "those commands back on the > stack", but I think we'd have to remember not a list of commands, but the > raw string of bytes from the start of the \while to the \endwhile (or > equivalent), because any psql vars within that block could themselves be a > non-parameter part of a command: > > -- this is how I fake an 'exit 0' now: > \set work_needs_to_be_done t > select > case > when :'work_needs_to_be_done'::boolean then '' > else '\q' > end as cmd > \gset > :cmd > > -- ridiculous example to illustrate complications in remembering past > commands > select > case > when random() < 0.5 then '\ir my_script.sql' > when random() < 0.7 'select * from a_table; select count(*) from > another_table;' > else 'select null as foo;' > end as cmd > \gset > :cmd > > And even then, things get complicated, because an \ir include which makes it > this iteration might not make it the next, and the \endwhile might have been > inside that include, or vice-versa, an included file starts a \while it > doesn't finish. I see your point. Just out of curiosity, why in the world don't you use something other than psql for scripting? I mean, if you accessed the data from Perl or Python or $INSERT_YOUR_FAVORITE_SCRIPTING_LANGUAGE_HERE, you could do all of this stuff very easily without any contortions. I've always thought of psql as something that's fine for interactive use and goofy trivial scripting but not really suitable for serious work. I grant that you seem to be making it serve the purpose, but, man. > So maybe what we store is a stack of buffers that are currently open (STDIN > being captured as a buffer only when a \while starts, everything else being > files), and additionally have a stack of positions where a \while started > (buffer_id, position in buffer). Yeah, sounds about right. > Additionally, we could assert that all \while-\endwhile pairs must happen in > the same MainLoop (aka file), and mismatches are an error. Sounds prudent. > I'm happy to keep sketching out what control structures might look like and > how to implement them. But I'm also happy to leave while/for loops out for > now. Right, I think that while/for can be left for another time, as long as the plan doesn't preclude doing it someday. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
2016-12-19 18:30 GMT+01:00 Robert Haas <robertmhaas@gmail.com>:
On Sat, Dec 17, 2016 at 3:39 PM, Corey Huinker <corey.huinker@gmail.com> wrote:
>> To implement \while, we'd also need to remember previous commands so
>> that when we reach the end of the loop, we can rewind and put all of
>> those commands back on the stack to be executed again, or perhaps to
>> be skipped if the \while condition turns out now to be false.
>
> This might be what you meant when you said "those commands back on the
> stack", but I think we'd have to remember not a list of commands, but the
> raw string of bytes from the start of the \while to the \endwhile (or
> equivalent), because any psql vars within that block could themselves be a
> non-parameter part of a command:
>
> -- this is how I fake an 'exit 0' now:
> \set work_needs_to_be_done t
> select
> case
> when :'work_needs_to_be_done'::boolean then ''
> else '\q'
> end as cmd
> \gset
> :cmd
>
> -- ridiculous example to illustrate complications in remembering past
> commands
> select
> case
> when random() < 0.5 then '\ir my_script.sql'
> when random() < 0.7 'select * from a_table; select count(*) from
> another_table;'
> else 'select null as foo;'
> end as cmd
> \gset
> :cmd
>
> And even then, things get complicated, because an \ir include which makes it
> this iteration might not make it the next, and the \endwhile might have been
> inside that include, or vice-versa, an included file starts a \while it
> doesn't finish.
I see your point. Just out of curiosity, why in the world don't you
use something other than psql for scripting? I mean, if you accessed
the data from Perl or Python or
$INSERT_YOUR_FAVORITE_SCRIPTING_LANGUAGE_HERE, you could do all of
this stuff very easily without any contortions. I've always thought
of psql as something that's fine for interactive use and goofy trivial
scripting but not really suitable for serious work. I grant that you
seem to be making it serve the purpose, but, man.
The integration of any scripting environment with SQL is much more less than in psql - just some easy scenarios are in psql natural.
It is similar why some people like me, prefer PLpgSQL against Java, Perl, ...
years ago there was a bash integrated with postgres - for me nice idea, but this project is dead.
> So maybe what we store is a stack of buffers that are currently open (STDIN
> being captured as a buffer only when a \while starts, everything else being
> files), and additionally have a stack of positions where a \while started
> (buffer_id, position in buffer).
Yeah, sounds about right.
> Additionally, we could assert that all \while-\endwhile pairs must happen in
> the same MainLoop (aka file), and mismatches are an error.
Sounds prudent.
> I'm happy to keep sketching out what control structures might look like and
> how to implement them. But I'm also happy to leave while/for loops out for
> now.
Right, I think that while/for can be left for another time, as long as
the plan doesn't preclude doing it someday.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
I see your point. Just out of curiosity, why in the world don't you
use something other than psql for scripting? I mean, if you accessed
the data from Perl or Python or
$INSERT_YOUR_FAVORITE_SCRIPTING_LANGUAGE_HERE, you could do all of
this stuff very easily without any contortions. I've always thought
of psql as something that's fine for interactive use and goofy trivial
scripting but not really suitable for serious work. I grant that you
seem to be making it serve the purpose, but, man.
I'm coming to the realization that this sentiment, when applied to my primary application, is probably correct...
In my situation the scripting language of choice is Bash - which largely acts as glue for programs such as psql, pdftk, enscript, and the R language.
Being able to do more conditional work in psql would make setting up more robust scripts easier and without either losing transaction capabilities or session pooling for improved performance when large numbers of small commands are run in between flow control in done in bash.
David J.
I see your point. Just out of curiosity, why in the world don't you>
> -- this is how I fake an 'exit 0' now:
> \set work_needs_to_be_done t
> select
> case
> when :'work_needs_to_be_done'::boolean then ''
> else '\q'
> end as cmd
> \gset
> :cmd
>
> -- ridiculous example to illustrate complications in remembering past
> commands
> select
> case
> when random() < 0.5 then '\ir my_script.sql'
> when random() < 0.7 'select * from a_table; select count(*) from
> another_table;'
> else 'select null as foo;'
> end as cmd
> \gset
> :cmd
>
> And even then, things get complicated, because an \ir include which makes it
> this iteration might not make it the next, and the \endwhile might have been
> inside that include, or vice-versa, an included file starts a \while it
> doesn't finish.
use something other than psql for scripting? I mean, if you accessed
the data from Perl or Python or
$INSERT_YOUR_FAVORITE_SCRIPTING_LANGUAGE_HERE, you could do all of
this stuff very easily without any contortions. I've always thought
of psql as something that's fine for interactive use and goofy trivial
scripting but not really suitable for serious work. I grant that you
seem to be making it serve the purpose, but, man.
Since you asked:
Heh. I *don't* do the second example, I was just pointing out that those things could be done, not that they should be done, and how hard it would be to implement loops when the source code is potentially coming from a stream.
My current client does use mostly python, but also perl, and ruby, and PHP and, reactjs, and $NEW_THING.
Here are the reasons I often prefer psql:
- Wiring up a python script to do one if-then in the middle of 40 SQL statements goes a long way toward obfuscating what SQL is going to be run.
- Wiring up a python script to do one if-then in the middle of 40 SQL statements goes a long way toward obfuscating what SQL is going to be run.
- Packaging up the SQL statements in a DO $$ $$; block conceals what statements were run, and how long they took.
- In python, etc, it's up to me to show rowcounts and timings.
- On very small docker-ish systems, the fewer things I have to install, the better, and golly, python is large these days.
- When doing work for regulated industry clients (SOX,HIPAA, school district PII, etc), the auditors like seeing clearly what SQL _will_ run, what SQL _did_ run, and what was affected. psql scripts with echo-queries set and captured output do that nicely. Installing extra scripting languages gives them the vapors, and now we need an auditor that thinks they know two languages, not one. I'm not saying it makes sense, I'm saying fewer dependencies gets a auditor's checkbox checked sooner.
Right, I think that while/for can be left for another time, as long as
the plan doesn't preclude doing it someday.
+1
David G. Johnston wrote: > Being able to do more conditional work in psql would make setting up more > robust scripts easier and without either losing transaction capabilities or > session pooling for improved performance when large numbers of small > commands are run in between flow control in done in bash. Have you tried to script processes in shell using a single background psql process with which the shell code communicates using a pipe? I've long been curious about that approach, but never had a strong need enough to actually write the code. It should be possible. -- Álvaro Herrera https://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
David G. Johnston wrote:
> Being able to do more conditional work in psql would make setting up more
> robust scripts easier and without either losing transaction capabilities or
> session pooling for improved performance when large numbers of small
> commands are run in between flow control in done in bash.
Have you tried to script processes in shell using a single background
psql process with which the shell code communicates using a pipe? I've
long been curious about that approach, but never had a strong need
enough to actually write the code. It should be possible.
I've envisioned and read up a bit on the approach but the cost-benefit hasn't yet made actually doing it worthwhile.
I do pretty much myself run all of the scripts I've been writing - the cost-benefit ratio is likely to change once they are turned over to a non-programmer to run.
David J.
Ok, so activity on this thread has died down. I'm not sure if that's consensus or exhaustion.
Are we good with:
Are we good with:
- implementing basic \if EXPR \elseif EXPR \else \endif, where the EXPR is an expression but is currently limited to a simple string that will be evaluated for truth via ParseVariableBool()?
- moving beyond trivial expressions in a later commit?
- leaving loops out for now?
- moving beyond trivial expressions in a later commit?
- leaving loops out for now?
Hi
2017-01-14 0:20 GMT+01:00 Corey Huinker <corey.huinker@gmail.com>:
+1
+1
Ok, so activity on this thread has died down. I'm not sure if that's consensus or exhaustion.
the original idea \quit_if is leaved? It is pity - it is common use case - and because we cannot to implement macros in psql, then can be very useful
Are we good with:- implementing basic \if EXPR \elseif EXPR \else \endif, where the EXPR is an expression but is currently limited to a simple string that will be evaluated for truth via ParseVariableBool()?
+1
- moving beyond trivial expressions in a later commit?
the expressions are in nice to have category - there can be a logic
if there is only a variable, check the variable; else eval on server and check the result.
- leaving loops out for now?
+1
Regards
Pavel
Pavel Stehule <pavel.stehule@gmail.com> writes:
> 2017-01-14 0:20 GMT+01:00 Corey Huinker <corey.huinker@gmail.com>:
>> - leaving loops out for now?
> +1
I'm just going to say one thing about that: some people will remember
that you can build a Turing machine with either conditionals+iteration
or conditionals+recursion. I wonder what depth of include-file nesting
psql can support, or whether we'll be able to fix it to optimize tail
recursion of an include file. Because somebody will be asking for that
if this is the toolset you give them.
regards, tom lane
PS: if I'm being too obscure for you, consider:
$ cat loop.sql
\if :x < 1000 \echo :x \set x :x + 1 \include loop.sql
\fi
$ psql --set x=0 -f loop.sql
Somebody is going to think of that workaround for not having loops, and
then whine about how psql runs out of file descriptors and/or stack.
> $ cat loop.sql > \if :x < 1000 > \echo :x > \set x :x + 1 > \include loop.sql > \fi > $ psql --set x=0 -f loop.sql Nice one! CPP does not have arithmetic, so it is harder to do that because one must reimplement arithmetic with #if... > Somebody is going to think of that workaround for not having loops, and > then whine about how psql runs out of file descriptors and/or stack. One can already have "include nested too deeply" errors, I guess, without a recursion. I would say that's this consequence is acceptable, and that this is a feature. I think having some kind of client-side test brings significant value because it would help writing application schema upgrades for instance, and that the this potential whining source is worth handling. -- Fabien.
On 1/13/17 11:22 PM, Tom Lane wrote: > I wonder what depth of include-file nesting > psql can support, or whether we'll be able to fix it to optimize tail > recursion of an include file. Because somebody will be asking for that > if this is the toolset you give them. I think the solution to that is straightforward: tell users that we hope to eventually support loops and that in the meantime if you try to work around that with recursion you get to keep both pieces when it breaks. While not ideal I think that's a lot better than throwing the whole idea out because some people will abuse it... -- Jim Nasby, Data Architect, Blue Treble Consulting, Austin TX Experts in Analytics, Data Architecture and PostgreSQL Data in Trouble? Get it in Treble! http://BlueTreble.com 855-TREBLE2 (855-873-2532)
On Sat, Jan 14, 2017 at 12:22 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > > $ cat loop.sql > \if :x < 1000 > \echo :x > \set x :x + 1 > \include loop.sql > \fi > $ psql --set x=0 -f loop.sql > > Somebody is going to think of that workaround for not having loops, and > then whine about how psql runs out of file descriptors and/or stack. Hmm, I think somebody just DID think of it. But personally this doesn't upset me a bit. If somebody complains about that particular thing, I think that would be an excellent time to suggest that they write a patch to add a looping construct. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
On Wed, Jan 18, 2017 at 3:24 AM, Robert Haas <robertmhaas@gmail.com> wrote: > On Sat, Jan 14, 2017 at 12:22 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote: >> >> $ cat loop.sql >> \if :x < 1000 >> \echo :x >> \set x :x + 1 >> \include loop.sql >> \fi >> $ psql --set x=0 -f loop.sql >> >> Somebody is going to think of that workaround for not having loops, and >> then whine about how psql runs out of file descriptors and/or stack. > > Hmm, I think somebody just DID think of it. > > But personally this doesn't upset me a bit. If somebody complains > about that particular thing, I think that would be an excellent time > to suggest that they write a patch to add a looping construct. Agreed. As far as I can see on this thread, something could be done, it is just that we don't know yet at which extent things could be done with the first shot. There are many things that could be done, but at least I'd suggest to get \if, \fi and \quit to satisfy the first requirements of this thread, and let loops out of it. I have switched the patch as "returned with feedback" as getting a new patch is going to require some thoughts to get the context handling done correctly on psql side. -- Michael
On Wed, Jan 18, 2017 at 12:08 AM, Michael Paquier <michael.paquier@gmail.com> wrote:
On Wed, Jan 18, 2017 at 3:24 AM, Robert Haas <robertmhaas@gmail.com> wrote:
> On Sat, Jan 14, 2017 at 12:22 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>
>> $ cat loop.sql
>> \if :x < 1000
>> \echo :x
>> \set x :x + 1
>> \include loop.sql
>> \fi
>> $ psql --set x=0 -f loop.sql
>>
>> Somebody is going to think of that workaround for not having loops, and
>> then whine about how psql runs out of file descriptors and/or stack.
>
> Hmm, I think somebody just DID think of it.
>
> But personally this doesn't upset me a bit. If somebody complains
> about that particular thing, I think that would be an excellent time
> to suggest that they write a patch to add a looping construct.
Agreed.
As far as I can see on this thread, something could be done, it is
just that we don't know yet at which extent things could be done with
the first shot. There are many things that could be done, but at least
I'd suggest to get \if, \fi and \quit to satisfy the first
requirements of this thread, and let loops out of it. I have switched
the patch as "returned with feedback" as getting a new patch is going
to require some thoughts to get the context handling done correctly on
psql side.
--
Michael
Fabien is pressed for time, so I've been speaking with him out-of-thread about how I should go about implementing it.
The v1 patch will be \if <expr>, \elseif <expr>, \else, \endif, where <expr> will be naively evaluated via ParseVariableBool().
The v1 patch will be \if <expr>, \elseif <expr>, \else, \endif, where <expr> will be naively evaluated via ParseVariableBool().
\ifs and \endifs must be in the same "file" (each MainLoop will start a new if-stack). This is partly for sanity (you can see the pairings unless the programmer is off in \gset meta-land), partly for ease of design (data structures live in MainLoop), but mostly because it would an absolute requirement if we ever got around to doing \while.
I hope to have something ready for the next commitfest.
As for the fate of \quit_if, I can see it both ways. On the one hand, it's super-simple, already written, and handy.
On the other hand, it's easily replaced by
\if <expr>\q\endif
So I'll leave that as a separate reviewable patch.
As for loops, I don't think anyone was pushing for implementing \while now, only to have a decision about what it would look like and how it would work. There's a whole lot of recording infrastructure (the input could be a stream) needed to make it happen. Moreover, I think \gexec scratched a lot of the itches that would have been solved via a psql looping structure.