Named Operators

Technically correct name of this feature would be Readable Names for
Operators, or Pronounceable Names for Operators. But I'd like to call
it Named Operators.

With this patch in place, the users can name the operators as
:some_pronounceable_name: instead of having to choose from the special
characters like #^&@. For example, users will be able to create and
use operators like:

select
expr1 :distance: expr2,
expr3 :contains_all: expr4,
expr5 :contains_any: expr6
expr7 :contains_exactly_two_of: expr8
from mytable;

instead of being forced to use these:

select
expr1 <#> expr2,
expr3 ?& expr4,
expr5 ?| expr6
expr7 ??!! expr8 -- ¯\_(ツ)_/¯
from mytable;

I think Named Operators will significantly improve the readability
of queries.

After a little trial-an-error, it was easy to develop the scan.l
rules to implement this feature, without flex barfing. The hard part
has been convincing myself that this is a safe implementation, even
though there are no regressions in `make check`. I am unsure of this
implementation's compatibility with the SQL Spec, and I'm not able to
envision problems with its interaction with some current or potential
feature of Postgres. So I'd really appreciate feedback from someone
who is conversant with the SQL Spec.

If the colon character being used as a delimiter poses a
challenge, other good candidates for the delimiter seem to be one of
~^` Although I haven't tested any of these to see if they cause a
regression. The colon character is be preferable for the delimiter,
since it is already used in the typecast :: operator.

I tried to strip the delimiters/colons from the name right in the
scanner, primarily because that would allow the identifier part of the
name to be as long as NAMEDATALEN-1, just like other identifiers
Postgres allows. Added benefit of stripping delimiters was that the
rest of the code, and catalogs/storage won't have to see the
delimiters. But stripping the delimiters made the code brittle; some
places in code now had to be taught different handling depending on
whether the operator name was coming from the user command, or from
the catalogs. I had to special-case code in pg_dump, as well. To share
code with frontends like pg_dump, I had to place code in src/common/.
I was still not able to address some obvious bugs.

By retaining the delimiters : in the name, the code became much
simpler; pg_dump support came for free! The bugs became a non-issue.
To see how much code and complexity was reduced, one can see this
commit [1]. The downside of retaining the delimiters is that the
identifier part of the name can be no more than NAMEDATALEN-3 in
length.

Because of the minimal changes to the scanner rules, and no
changes in the grammar, I don't think there's any impact on precedence
and associativity rules of the operators. I'd be happy to learn
otherwise.

Here's a rudimentary test case to demonstrate the feature:

create operator :add_point: (function = box_add, leftarg = box,
rightarg = point);
create table test(a box);
insert into test values('((0,0),(1,1))'), ('((0,0),(2,1))');
select a as original, a :add_point: '(1,1)' as modified from test;
drop operator :add_point:(box, point);

Feedback will be much appreciated!

[1]: Commit: Don't strip the delimiters
https://github.com/gurjeet/postgres/commit/62d11a578e5325c32109bb2a55a624d0d89d4b7e

[2]: Git branch named_operators
https://github.com/gurjeet/postgres/tree/named_operators

Best regards,
Gurjeet
http://Gurje.et