documentation synopsis grammar

On Friday, May 11, 2018, Peter Korim <pkorim@gmail.com> wrote:

Hallo

I search for it and did notsucceeded.
Where is exact definition of the grammar used here?

I thing the text "[,...]" is unclear concerning what is to be repeted.

Not that I've found, do you have an example?

here:
*https://www.postgresql.org/docs/9.6/static/notation.html
<https://www.postgresql.org/docs/9.6/static/notation.html&gt;*
you state
*" Dots (...) mean that the preceding element can be repeated. "*

from that definition the fragment "[,...]" means zero or more commas.

It is often so that not one element but several elements are repeated .

In which case each one requires a comma separating it from the previous one.

Sometimes from the context, one may find out how it is meant.

In other cases it not so clear.

Again, You will need to point out an example. But not all options that
allow multiples require commas to separate them.

Well do you have a definition or third party standard?

You linked to it above...

David J.

On Friday, May 11, 2018, David G. Johnston <david.g.johnston@gmail.com>
wrote:

On Friday, May 11, 2018, Peter Korim <pkorim@gmail.com> wrote:

Hallo

I search for it and did notsucceeded.
Where is exact definition of the grammar used here?

I thing the text "[,...]" is unclear concerning what is to be repeted.

Not that I've found, do you have an example?

Ok, I think I get your confusion now...

ANALYZE [ VERBOSE ] [ *table_name* [ ( *column_name* [, ...] ) ] ]

column_name can be repeated. If it is each name is separated from the
previous one by a comma. The brackets indicate optional repetition of the
immediately preceding element. The presence or absence of a comma in the
bracket indicates whether each option repeat needs a comma separator. It's
not the comma that is being repeated.

David J.

Please keep communications on-list.

On Fri, May 11, 2018 at 3:05 PM, Peter Korim <pkorim@gmail.com> wrote:

Ok David

by your definition and following synopsis:
CREATE [ OR REPLACE ] FUNCTION
name (
​​
[ [ argmode ] [ argname ] argtype [ { DEFAULT | = } default_expr ] [, ...]
] )
(seen here: https://www.postgresql.org/docs/9.6/static/sql-
createfunction.html)

should be valid this:
CREATE [ OR REPLACE ] FUNCTION
name ( argtype = default_expr , = default_expr , =
default_expr )

​But it obviously isn't...
​
I suppose the [, ...] could be placed after the "]" to its right...but the
only mandatory element in that [] block is "argtype" so the [, ...] refers
to it (or rather, everything in between the [ ] of which it is a part.

(argtype, argtype, argtype)

Each one of those argtype values can be surrounded by any of the optional
elements surrounding it in the syntax. The whole part between the ( )
being optional, hence the idiomatic ( [ ] ) syntax.

These are written by humans for humans - and in some cases do take a bit of
thinking (and, sometimes, reading the descriptions about the items in
question) about what ultimately makes sense to understanding exactly what
might or might not be valid.

I'm suspecting that our best bet is leave the notation page a bit vague and
just clear up confusion when it arises. The example above, while probably
technically incorrect, is, I'm reasonably certain, common and saying its
wrong and fixing it is unlikely to happen given the rarity of questions
like this. Trying to describe exactly how that [, ...] works in the
presence of mandatory and optional parts is likely to add, not reduce,
confusion. As its stands "..." means repetition is possible and you, the,
reader, are assumed to figure out what exactly is repeatable. You'll
likely get a syntax error quickly if you guess wrong and these lists are
very responsive if you get stumped.

You are welcome to suggest corrections to the documentation, and even
better, submit patches, if you find the existing conventions/notations page
unconscionably lacking.

David J.

"David G. Johnston" <david.g.johnston@gmail.com> writes:

I'm suspecting that our best bet is leave the notation page a bit vague and
just clear up confusion when it arises. The example above, while probably
technically incorrect, is, I'm reasonably certain, common and saying its
wrong and fixing it is unlikely to happen given the rarity of questions
like this.

Yeah; a quick grep suggests that there are several hundred occurrences
of this notation in our reference pages alone. Even if somebody were
initially confused, they'd soon figure it out, I should think. Certainly
we've had few if any complaints about this point before.

The bigger question though is, if we don't like this notation, what
notation would we replace it with? We could be formally correct by
rewriting all of these syntax synopses in BNF, but I think most people
are not terribly familiar with that and would be more confused, not less.
Our actual bison grammar, which is BNF-equivalent I think, is certainly
arcane enough to scare off non-experts.

There was a related discussion recently:

/messages/by-id/152110913499.1223.7026776990975251345@wrigleys.postgresql.org

The problems discussed there with our description of set-operation syntax
are really a lot worse than this issue, I think. And yet we still opted
not to change the documentation, because it seemed that anything that's
more formally correct would also be a lot more incomprehensible.

I don't want to sound like I think what we've got now is the peak of
perfection, because it isn't. But we have to strike a balance between
formal correctness and readability for users who aren't familiar with
formal syntax notations. It's a difficult problem.

regards, tom lane

"David G. Johnston" <david.g.johnston@gmail.com> writes:

I'm suspecting that our best bet is leave the notation page a bit vague

and

just clear up confusion when it arises. The example above, while

probably

technically incorrect, is, I'm reasonably certain, common and saying its
wrong and fixing it is unlikely to happen given the rarity of questions
like this.

Yeah; a quick grep suggests that there are several hundred occurrences
of this notation in our reference pages alone. Even if somebody were
initially confused, they'd soon figure it out, I should think. Certainly
we've had few if any complaints about this point before.

I understand the motivation .
This notation considerably shorten the synopsis.
In many cases it is shiny clear what is the correct grammar.
everybody should know what is repeating in argument list like here:
( [ [ argmode ] [ argname ] argtype [ { DEFAULT | = }
default_expr ] [, ...] ] )
otherwise it is not possible to write any function

the same is with return clause (I hope ):
( column_name column_type [, ...] )
though it is less ofently use .

the problem may arise with with rarely use and/or new features like e.g.:

create table exclude clause (does anybody ude it ? :) no use of construct
= no complaints)
EXCLUDE [ USING index_method ] ( exclude_element WITH operator [,
... ] ) index_parameters [ WHERE ( predicate ) ]
if I do not consult bison grammar I bet the repeating item comma
separated group is
exclude_element WITH operator

so the description would be everything at the level of parents (either
round, curly or square) same as "[,... ]" repeats.
NOTE THAT: some braces are tokens, other are metasymbols.
On the other hand there is no such thing as "obligatory repetition". So
either the whole list is obligatory or optional.
This can be expressed in adding braces around the list.

so there is no need to use ambiguous symbols [,...] [...] ... ,...

Last but not least:
When you say the synopsis is for humans (excuse to not to be exact?) .
ok no problem.
Question is where in documentation is formal syntax definition of language?
gram.y?

What is my actual motivation is to create ANTLR4 grammar . The synopsis is
better starting point than bison . You simply replace parents by tokens.
next curly braces by parents next you determine what is to be repeated in
[,...] constructs and put it into the XYZ(,XYZ )? rule and finally
replace square barces by ()? construct.

with best wishes Peter

The bigger question though is, if we don't like this notation, what
notation would we replace it with? We could be formally correct by
rewriting all of these syntax synopses in BNF, but I think most people
are not terribly familiar with that and would be more confused, not less.
Our actual bison grammar, which is BNF-equivalent I think, is certainly
arcane enough to scare off non-experts.

There was a related discussion recently:

/messages/by-id/152110913499.1223.7026776990975251345@wrigleys.postgresql.org

The problems discussed there with our description of set-operation syntax
are really a lot worse than this issue, I think. And yet we still opted
not to change the documentation, because it seemed that anything that's
more formally correct would also be a lot more incomprehensible.

I don't want to sound like I think what we've got now is the peak of
perfection, because it isn't. But we have to strike a balance between
formal correctness and readability for users who aren't familiar with
formal syntax notations. It's a difficult problem.

regards, tom lane

Hallo Tom

so at first the new explanation is (and should be written in definition of
notation):

On May 11, 2018, at 8:15 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

"David G. Johnston" <david.g.johnston@gmail.com> writes:

I'm suspecting that our best bet is leave the notation page a bit vague and
just clear up confusion when it arises. The example above, while probably
technically incorrect, is, I'm reasonably certain, common and saying its
wrong and fixing it is unlikely to happen given the rarity of questions
like this.

Yeah; a quick grep suggests that there are several hundred occurrences
of this notation in our reference pages alone. Even if somebody were
initially confused, they'd soon figure it out, I should think. Certainly
we've had few if any complaints about this point before.

The bigger question though is, if we don't like this notation, what
notation would we replace it with? We could be formally correct by
rewriting all of these syntax synopses in BNF, but I think most people
are not terribly familiar with that and would be more confused, not less.
Our actual bison grammar, which is BNF-equivalent I think, is certainly
arcane enough to scare off non-experts.

There was a related discussion recently:

/messages/by-id/152110913499.1223.7026776990975251345@wrigleys.postgresql.org

The problems discussed there with our description of set-operation syntax
are really a lot worse than this issue, I think. And yet we still opted
not to change the documentation, because it seemed that anything that's
more formally correct would also be a lot more incomprehensible.

I don't want to sound like I think what we've got now is the peak of
perfection, because it isn't. But we have to strike a balance between
formal correctness and readability for users who aren't familiar with
formal syntax notations. It's a difficult problem.

Perhaps a way around it is having more practical examples that highlight
the way the language can be used. Even with an understanding of the
PostgreSQL, let alone SQL, syntax, I find that I continue to learn things
the language can do even to this day when I see an example. Sometimes
the grammar masks a lot of the power :-)

I would think changing the grammar at this point would cause even more
confusion, but more examples to capture the power should shed more
light on how to do things.

Jonathan