At Wed, 28 Nov 2018 14:41:40 +0900, Tatsuro Yamada <yamada.tats...@lab.ntt.co.jp> wrote in
<54bd214b-d0d3-8654->>>>
* tab_completion_alter_index_set_statistics.patch
=======
There are two problems. You can use these DDL before testing.
#create table hoge (a integer, b integer);
#create index ind_hoge on hoge (a, (a + b), (a * b));
1) Can't get column names
# alter index ind_hoge alter column <tab!><tab!>... but can't
# complete.
Currently the only continueable rule to the rule is SET
STATISTICS so we usually expect the number of an expression
column there. Even though we actually name every expression
column in an index, users hardly see the names. The names are in
the index column number order in your example, but what if the
name of the first column were 'foo'?
=# alter index ind_hoge2 alter column
expr expr1 foo
We could still *guess* what is expr or exrp1 but I don't think it
helps much. (Note: foo is not usable in this context as it's a
non-expression column.)
Thanks for your comment.
We can get column name by using "\d index_name" like this:
# \d ind_hoge
Index "public.ind_hoge"
Column | Type | Key? | Definition
--------+---------+------+------------
a | integer | yes | a
expr | integer | yes | (a + b)
expr1 | integer | yes | (a * b)
btree, for table "public.hoge"
So, I suppose that it's easy to understand what column is an
expression column.
Yeah, actually we can find them there, but I don't think it's
popular. I suppose that most of people are unconcious of the
names since they didn't named them. Moreover what makes me
uneasy with this is that it doesn't suggest attribute numbers,
which are firstly expected there. But anyway it is impossible
with readline since suggestions are sorted as strings. ("1",
"11", "12", "2" order, I don't think indexes often have that many
columns, though.)
Okay, I understand.
- For now, there is no column_number column on "\d index_name" result.
So, if tab-completion suggested column_numbers, user can't match these
easily.
- Suggested column_name and column_number are sorted as a string by readline.
These are an unnatural. But this is another topic on this thread.
Example:
# alter index ind_hoge alter column <tab!>
c expr expr1 expr10 expr11 expr2 expr3 expr4 expr5
expr6 expr7 expr8 expr9
Of course, user will get syntax error if user chose "a" column like a
"foo" which is
non-expression column as you mentioned.
Probably, I will be able to fix the patch to get only expression
columns from the index.
Should I do that?
Nope. That's just too-much. We are already showing unusable
suggestions in other places, for example, exiting tables for
CREATE TABLE. (It is useful for me as it suggests what should be
placed there.)
I see.
I prefer to use column name instead column number because
there is no column number on \d index_name and \d+ index_name.
Some people prefer names even if they are implicitly
given. Others (including myself) prefer numbers.
So, we should better to vote to decide implementation of the tab-completion.
Which is better to use either column_names or column_numbers?
I'd like to hear opinions from others. :)
2) I expected column names for column numbers after "SET STATISTICS",
but
tab-completion gave schema names
# alter index ind_hoge alter column expr SET STATISTICS <tab!>
information_schema. pg_catalog. pg_temp_1. pg_toast.
pg_toast_temp_1. public.
This is the result of STATISTICS <things> completion. SET
STATISTICS always doesn't take statistics name so this is safe.
:)
So I think it is reasonable.
Thanks.
Regards,
Tatsuro Yamada
