Tab completion for CREATE SCHEMAAUTHORIZATION

ilmari@ilmari.org (Dagfinn Ilmari Mannsåker) writes:

Hi Hackers,

I just noticed there's no tab completion for CREATE SCHEMA
AUTHORIZATION, nor for anything after CREATE SCHEMA <name>.

Please find attached a patch that adds this.

Added to the 2021-09 commit fest: https://commitfest.postgresql.org/34/3252/

- ilmari

ilmari@ilmari.org (Dagfinn Ilmari Mannsåker) writes:

ilmari@ilmari.org (Dagfinn Ilmari Mannsåker) writes:

Hi Hackers,

I just noticed there's no tab completion for CREATE SCHEMA
AUTHORIZATION, nor for anything after CREATE SCHEMA <name>.

Please find attached a patch that adds this.

Added to the 2021-09 commit fest: https://commitfest.postgresql.org/34/3252/

Here's an updated version that also reduces the duplication between the
various role list queries.

- ilmari

Hello Dagfinn,

I had a look at your patch and below are my review comments.
Please correct me if I am missing something.

1. For me the patch does not apply cleanly. I have been facing the error
of trailing whitespaces.
surajkhamkar@localhost:postgres$ git apply
v2-0001-Add-tab-completion-for-CREATE-SCHEMA.patch
v2-0001-Add-tab-completion-for-CREATE-SCHEMA.patch:25: trailing
whitespace.
#define Query_for_list_of_schema_roles \
v2-0001-Add-tab-completion-for-CREATE-SCHEMA.patch:26: trailing
whitespace.
Query_for_list_of_roles \
v2-0001-Add-tab-completion-for-CREATE-SCHEMA.patch:30: trailing
whitespace.
#define Query_for_list_of_grant_roles \
v2-0001-Add-tab-completion-for-CREATE-SCHEMA.patch:31: trailing
whitespace.
Query_for_list_of_schema_roles \
v2-0001-Add-tab-completion-for-CREATE-SCHEMA.patch:32: trailing
whitespace.
" UNION ALL SELECT 'PUBLIC'"\
error: patch failed: src/bin/psql/tab-complete.c:758
error: src/bin/psql/tab-complete.c: patch does not apply

2. We can remove space in before \ and below change
+" UNION ALL SELECT 'PUBLIC'" \

Should be,
+" UNION ALL SELECT 'PUBLIC' "\

3. role_specification has CURRENT_ROLE, CURRENT_USER and SESSION_USER.
But current changes are missing CURRENT_ROLE.
postgres@53724=#CREATE SCHEMA AUTHORIZATION
CURRENT_USER pg_execute_server_program pg_read_all_data

pg_read_all_stats pg_signal_backend pg_write_all_data

SESSION_USER pg_database_owner pg_monitor

pg_read_all_settings pg_read_server_files
pg_stat_scan_tables
pg_write_server_files surajkhamkar

4. I'm not sure about this but do we need to enable tab completion for IF
NOT EXIST?

5. I think we are not handling IF NOT EXIST that's why it's not
completing tab completion
for AUTHORIZATION.

6. As we are here we can also enable missing tab completion for ALTER
SCHEMA.
After OWNER TO we should also get CURRENT_ROLE, CURRENT_USER and
SESSION_USER.
postgres@53724=#ALTER SCHEMA sch owner to
pg_database_owner pg_monitor
pg_read_all_settings
pg_read_server_files pg_stat_scan_tables
pg_write_server_files
pg_execute_server_program pg_read_all_data pg_read_all_stats

pg_signal_backend pg_write_all_data surajkhamkar

7. Similarly, as we can drop multiple schemas' simultaneously, we should
enable tab completion for
comma with CASCADE and RESTRICT
postgres@53724=#DROP SCHEMA sch
CASCADE RESTRICT

Thanks.

On Sun, Aug 8, 2021 at 2:39 AM Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
wrote:

Hi Suraj,

Suraj Khamkar <khamkarsuraj.b@gmail.com> writes:

Hello Dagfinn,

I had a look at your patch and below are my review comments.
Please correct me if I am missing something.

1. For me the patch does not apply cleanly. I have been facing the error
of trailing whitespaces.

I do not get these errors, neither with the patch file I still have
locally, or by saving the attachment from my original email. Are you
sure something in your download process hasn't converted it to Windows
line endings (\r\n), or otherwise mangled the whitespace?

2. We can remove space in before \ and below change
+" UNION ALL SELECT 'PUBLIC'" \

Should be,
+" UNION ALL SELECT 'PUBLIC' "\

The patch doesn't add any lines that end with quote-space-backslash.
As for the space before the quote, that is not necessary either, since the
next line starts with a space after the quote. Either way, the updated
version of the patch doesn't add any new lines with continuation after a
string constant, so the point is moot.

3. role_specification has CURRENT_ROLE, CURRENT_USER and SESSION_USER.
But current changes are missing CURRENT_ROLE.

Ah, I was looking at the documentation for 13, but CURRENT_ROLE is only
allowed in this context as of 14. Fixed.

4. I'm not sure about this but do we need to enable tab completion for IF
NOT EXIST?

5. I think we are not handling IF NOT EXIST that's why it's not
completing tab completion
for AUTHORIZATION.

As you note, psql currently doesn't currently tab-complete IF NOT EXISTS
for any command, so that would be a subject for a separate patch.

6. As we are here we can also enable missing tab completion for ALTER
SCHEMA.
After OWNER TO we should also get CURRENT_ROLE, CURRENT_USER and
SESSION_USER.

I did an audit of all the uses of Query_for_list_of_roles, and there
turned out be several more that accept CURRENT_ROLE, CURRENT_USER and
SESSION_USER that they weren't tab-completed for. I also renamed the
constant to Query_for_list_of_owner_roles, but I'm not 100% happy with
that name either.

7. Similarly, as we can drop multiple schemas' simultaneously, we should
enable tab completion for
comma with CASCADE and RESTRICT
postgres@53724=#DROP SCHEMA sch
CASCADE RESTRICT

The tab completion code for DROP is generic for all object types (see
the words_after_create array and the create_or_drop_command_generator
function), so that should be done genericallly, and is thus outside the
scope for this patch.

Thanks.

Thanks for the review. Updated patch attached, with the CURRENT/SESSION
ROLE/USER changes for other commands separated out.

- ilmari

On Mon, Aug 09, 2021 at 07:00:02PM +0100, Dagfinn Ilmari Mannsåker wrote:

Thanks for the review. Updated patch attached, with the CURRENT/SESSION
ROLE/USER changes for other commands separated out.

+#define Query_for_list_of_owner_roles \
+Query_for_list_of_roles \
" UNION ALL SELECT 'CURRENT_ROLE'"\
" UNION ALL SELECT 'CURRENT_USER'"\
" UNION ALL SELECT 'SESSION_USER'"
I don't object to the refactoring you are doing here with three
Query_for_list_of_*_roles each one depending on the other for clarity.
Neither do I really object to not using COMPLETE_WITH_QUERY() with
some extra UNION ALL hardcoded in each code path as there 6 cases for
_owner_, 6 for _grant_ and 6 for _roles if my count is right. Still,
if I may ask, wouldn't it be better to document a bit what's the
expectation behind each one of them? Perhaps the names of the queries
are too generic for the purposes where they are used (say _grant_ for
CREATE USER MAPPING looks confusing)?

+       else if (Matches("CREATE", "SCHEMA", "AUTHORIZATION"))
+               COMPLETE_WITH_QUERY(Query_for_list_of_owner_roles);
+       else if (Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION"))
+               COMPLETE_WITH_QUERY(Query_for_list_of_owner_roles);
+       else if (Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION", MatchAny))
+               COMPLETE_WITH("CREATE", "GRANT");
+       else if (Matches("CREATE", "SCHEMA", MatchAny))
+               COMPLETE_WITH("AUTHORIZATION", "CREATE", "GRANT");
Looks like you forgot the case "CREATE SCHEMA AUTHORIZATION MatchAny"
that should be completed by GRANT and CREATE.
--
Michael

Thanks Dagfinn for the updated patches.

I do not get these errors, neither with the patch file I still have

locally, or by saving the attachment from my original email. Are you
sure something in your download process hasn't converted it to Windows
line endings (\r\n), or otherwise mangled the whitespace?

No. I have downloaded patch on linux and patch also doesn't have any
trailing
spaces, though it is throwing an error.

Here are few other comments,

1. USER MAPPING does not have SESSION_USER as username in syntax
(though it works) but your changes provide the same. Also, we have USER
in
list which is missing in current code changes.
CREATE USER MAPPING [ IF NOT EXISTS ] FOR { user_name | USER |
CURRENT_ROLE | CURRENT_USER | PUBLIC }
SERVER server_name
[ OPTIONS ( option 'value' [ , ... ] ) ]
2. It might not be a scope of this ticket but as we are changing the
query for ALTER GROUP,
we should complete the role_specification after ALTER GROUP.
postgres@17077=#ALTER GROUP
pg_database_owner pg_monitor
pg_read_all_settings
pg_read_server_files pg_stat_scan_tables
pg_write_server_files
surajkhamkar. pg_execute_server_program pg_read_all_data

pg_read_all_stats pg_signal_backend pg_write_all_data

3. Missing schema_elements after CREATE SCHEMA AUTHORIZATION username to
tab-complete .
schema_elements might be CREATE, GRAND etc.

Thanks..

On Mon, Aug 9, 2021 at 11:30 PM Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
wrote:

On Wed, Aug 11, 2021 at 10:16:15AM +0900, Michael Paquier wrote:

+       else if (Matches("CREATE", "SCHEMA", "AUTHORIZATION"))
+               COMPLETE_WITH_QUERY(Query_for_list_of_owner_roles);
+       else if (Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION"))
+               COMPLETE_WITH_QUERY(Query_for_list_of_owner_roles);
+       else if (Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION", MatchAny))
+               COMPLETE_WITH("CREATE", "GRANT");
+       else if (Matches("CREATE", "SCHEMA", MatchAny))
+               COMPLETE_WITH("AUTHORIZATION", "CREATE", "GRANT");
Looks like you forgot the case "CREATE SCHEMA AUTHORIZATION MatchAny"
that should be completed by GRANT and CREATE.

This patch has been waiting on author for more than a couple of weeks,
so I have marked it as returned with feedback in the CF app. Please
feel free to resubmit if you are able to work more on that.
--
Michael

Michael Paquier <michael@paquier.xyz> writes:

On Wed, Aug 11, 2021 at 10:16:15AM +0900, Michael Paquier wrote:

+       else if (Matches("CREATE", "SCHEMA", "AUTHORIZATION"))
+               COMPLETE_WITH_QUERY(Query_for_list_of_owner_roles);
+       else if (Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION"))
+               COMPLETE_WITH_QUERY(Query_for_list_of_owner_roles);
+       else if (Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION", MatchAny))
+               COMPLETE_WITH("CREATE", "GRANT");
+       else if (Matches("CREATE", "SCHEMA", MatchAny))
+               COMPLETE_WITH("AUTHORIZATION", "CREATE", "GRANT");
Looks like you forgot the case "CREATE SCHEMA AUTHORIZATION MatchAny"
that should be completed by GRANT and CREATE.

This patch has been waiting on author for more than a couple of weeks,
so I have marked it as returned with feedback in the CF app. Please
feel free to resubmit if you are able to work more on that.

Looks like I completely dropped the ball on this one, sorry. Here's a
rebased patch which uses the new COMPLETE_WITH_QUERY_PLUS functionality
added in commit 02b8048ba5dc36238f3e7c3c58c5946220298d71.

- ilmari

On Fri, Apr 14, 2023 at 05:04:35PM +0100, Dagfinn Ilmari Mannsåker wrote:

Looks like I completely dropped the ball on this one, sorry.

So did I ;)

Here's a
rebased patch which uses the new COMPLETE_WITH_QUERY_PLUS functionality
added in commit 02b8048ba5dc36238f3e7c3c58c5946220298d71.

Thanks, I'll look at it.
--
Michael

On Sat, Apr 15, 2023 at 11:06:25AM +0900, Michael Paquier wrote:

Thanks, I'll look at it.

+   else if (Matches("CREATE", "SCHEMA", "AUTHORIZATION", MatchAny) ||
+            Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION", MatchAny))
+       COMPLETE_WITH("CREATE", "GRANT");
+   else if (Matches("CREATE", "SCHEMA", MatchAny))
+       COMPLETE_WITH("AUTHORIZATION", "CREATE", "GRANT");

I had this grammar under my eyes a few days ago for a different patch,
and there are much more objects types that can be appended to a CREATE
SCHEMA, like triggers, sequences, tables or views, so this is
incomplete, isn't it?
--
Michael

Michael Paquier <michael@paquier.xyz> writes:

On Sat, Apr 15, 2023 at 11:06:25AM +0900, Michael Paquier wrote:

Thanks, I'll look at it.

+   else if (Matches("CREATE", "SCHEMA", "AUTHORIZATION", MatchAny) ||
+            Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION", MatchAny))
+       COMPLETE_WITH("CREATE", "GRANT");
+   else if (Matches("CREATE", "SCHEMA", MatchAny))
+       COMPLETE_WITH("AUTHORIZATION", "CREATE", "GRANT");

I had this grammar under my eyes a few days ago for a different patch,
and there are much more objects types that can be appended to a CREATE
SCHEMA, like triggers, sequences, tables or views, so this is
incomplete, isn't it?

This is for completing the word CREATE itself after CREATE SCHEMA
[[<name>] AUTHORIZATION] <name>. The things that can come after that
are already handled generically earlier in the function:

/* CREATE */
/* complete with something you can create */
else if (TailMatches("CREATE"))
matches = rl_completion_matches(text, create_command_generator);

create_command_generator uses the words_after_create array, which lists
all the things that can be created.

- ilmari

Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> writes:

Michael Paquier <michael@paquier.xyz> writes:

On Sat, Apr 15, 2023 at 11:06:25AM +0900, Michael Paquier wrote:

Thanks, I'll look at it.

+   else if (Matches("CREATE", "SCHEMA", "AUTHORIZATION", MatchAny) ||
+            Matches("CREATE", "SCHEMA", MatchAny, "AUTHORIZATION", MatchAny))
+       COMPLETE_WITH("CREATE", "GRANT");
+   else if (Matches("CREATE", "SCHEMA", MatchAny))
+       COMPLETE_WITH("AUTHORIZATION", "CREATE", "GRANT");

I had this grammar under my eyes a few days ago for a different patch,
and there are much more objects types that can be appended to a CREATE
SCHEMA, like triggers, sequences, tables or views, so this is
incomplete, isn't it?

This is for completing the word CREATE itself after CREATE SCHEMA
[[<name>] AUTHORIZATION] <name>. The things that can come after that
are already handled generically earlier in the function:

/* CREATE */
/* complete with something you can create */
else if (TailMatches("CREATE"))
matches = rl_completion_matches(text, create_command_generator);

create_command_generator uses the words_after_create array, which lists
all the things that can be created.

But, looking closer at the docs, only tables, views, indexes, sequences
and triggers can be created as part of a CREATE SCHEMA statement. Maybe
we should add a HeadMatches("CREATE", "SCHEMA") exception in the above?

- ilmari

On Tue, May 02, 2023 at 01:19:49PM +0100, Dagfinn Ilmari Mannsåker wrote:

Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> writes:

This is for completing the word CREATE itself after CREATE SCHEMA
[[<name>] AUTHORIZATION] <name>. The things that can come after that
are already handled generically earlier in the function:

/* CREATE */
/* complete with something you can create */
else if (TailMatches("CREATE"))
matches = rl_completion_matches(text, create_command_generator);

create_command_generator uses the words_after_create array, which lists
all the things that can be created.

You are right. I have completely forgotten that this code path would
append everything that supports CREATE for a CREATE SCHEMA command :)

But, looking closer at the docs, only tables, views, indexes, sequences
and triggers can be created as part of a CREATE SCHEMA statement. Maybe
we should add a HeadMatches("CREATE", "SCHEMA") exception in the above?

Yes, it looks like we are going to need an exception and append only
the keywords that are supported, or we will end up recommending mostly
things that are not accepted by the parser.
--
Michael

Michael Paquier <michael@paquier.xyz> writes:

On Tue, May 02, 2023 at 01:19:49PM +0100, Dagfinn Ilmari Mannsåker wrote:

Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> writes:

This is for completing the word CREATE itself after CREATE SCHEMA
[[<name>] AUTHORIZATION] <name>. The things that can come after that
are already handled generically earlier in the function:

/* CREATE */
/* complete with something you can create */
else if (TailMatches("CREATE"))
matches = rl_completion_matches(text, create_command_generator);

create_command_generator uses the words_after_create array, which lists
all the things that can be created.

You are right. I have completely forgotten that this code path would
append everything that supports CREATE for a CREATE SCHEMA command :)

But, looking closer at the docs, only tables, views, indexes, sequences
and triggers can be created as part of a CREATE SCHEMA statement. Maybe
we should add a HeadMatches("CREATE", "SCHEMA") exception in the above?

Yes, it looks like we are going to need an exception and append only
the keywords that are supported, or we will end up recommending mostly
things that are not accepted by the parser.

Here's an updated v3 patch with that. While adding that, I noticed that
CREATE UNLOGGED only tab-completes TABLE and MATERIALIZED VIEW, not
SEQUENCE, so I added that (and removed MATERIALIZED VIEW when part of
CREATE SCHEMA).

- ilmari

On Mon, May 08, 2023 at 05:36:27PM +0100, Dagfinn Ilmari Mannsåker wrote:

Here's an updated v3 patch with that. While adding that, I noticed that
CREATE UNLOGGED only tab-completes TABLE and MATERIALIZED VIEW, not
SEQUENCE, so I added that (and removed MATERIALIZED VIEW when part of
CREATE SCHEMA).

+       /* but not MATVIEW in CREATE SCHEMA */
+       if (HeadMatches("CREATE", "SCHEMA"))
+           COMPLETE_WITH("TABLE", "SEQUENCE");
+       else
+           COMPLETE_WITH("TABLE", "SEQUENCE", "MATERIALIZED VIEW");

This may look strange at first glance, but the grammar is what it
is.. Perhaps matviews could be part of that at some point. Or not. :)

+       /* only some object types can be created as part of CREATE SCHEMA */
+       if (HeadMatches("CREATE", "SCHEMA"))
+           COMPLETE_WITH("TABLE", "VIEW", "INDEX", "SEQUENCE", "TRIGGER",
+                         /* for INDEX and TABLE/SEQUENCE, respectively */
+                         "UNIQUE", "UNLOGGED");

Not including TEMPORARY is OK here as the grammar does not allow a
directly to create a temporary schema. The (many) code paths that
have TailMatches() to cope with CREATE SCHEMA would continue the
completion of added, but at least this approach avoids the
recommendation if possible.

That looks pretty much OK to me. One tiny comment I have is that this
lacks brackets for the inner blocks, so I have added some in the v4
attached.
--
Michael

On Tue, May 09, 2023 at 12:26:16PM +0900, Michael Paquier wrote:

That looks pretty much OK to me. One tiny comment I have is that this
lacks brackets for the inner blocks, so I have added some in the v4
attached.

The indentation was a bit wrong, so fixed it, and applied on HEAD.
--
Michael

Michael Paquier <michael@paquier.xyz> writes:

On Tue, May 09, 2023 at 12:26:16PM +0900, Michael Paquier wrote:

That looks pretty much OK to me. One tiny comment I have is that this
lacks brackets for the inner blocks, so I have added some in the v4
attached.

The indentation was a bit wrong, so fixed it, and applied on HEAD.

Thanks!

- ilmari