[utkarash2991]

Summary

Two things, both prerequisites for cutting a 0.2.1 release with the fix in PSQLEXT-1773:

1. Force CREATE EXTENSION through the full upgrade chain, per the suggestion in Schema-qualify hypopg_reset() and hypopg_get_indexdef() calls #12. Previously a fresh CREATE EXTENSION installed straight from a standalone "latest version" script, while ALTER EXTENSION UPDATE on an existing install walked a separate chain of incremental scripts - the two paths were only kept in sync by hand.

Now only 0.1.0 is a standalone base script; every later version (0.1.1, 0.1.2, 0.2.0) is reachable only via incremental upgrade scripts . CREATE EXTENSION with no version and ALTER EXTENSION UPDATE on an existing install both resolve through the same chain now - one code path, not two.

0.1.2->0.2.0 needs an explicit DROP FUNCTION first since the return signature changed (added the errors column) - confirmed this is actually required via the real Postgres error it produces without it: cannot change return type of existing function, HINT: Use DROP FUNCTION index_advisor(text) first.

2. Fixed CI entirely

Test plan

Added two tests:

• create_extension_default_version: unversioned CREATE EXTENSION resolves through the full chain to default_version.
• upgrade_from_oldest_version: a real ALTER EXTENSION UPDATE from 0.1.0 walks the whole chain and the function works correctly afterward - not just a version-pinned fresh install.

Tracked in https://linear.app/supabase/issue/PSQLEXT-1773

[AndrewJackson2020]

Would be nice to keep these in a different PR as they are totally independent.

[utkarash2991]

yeah ideally it should be in different PRs. Will be keeping in mind next time

[AndrewJackson2020]

did you create a branch both in this repo and your fork? looks like this one is only pointing at the fork. Not sure what the difference is between them but please make sure the fork in this repo is removed after this PR.

aj@Andrews-MacBook-Pro:~/Repos/mono/external/repos/supabase/index_advisor/main/ > git branch -vr | grep upg
rade-chain
  origin/utkarashsingh/psqlext-1773-force-upgrade-chain                   0bc39fe Fix CI and add PG17 to the test matrix
  utkarash_fork/utkarashsingh/psqlext-1773-force-upgrade-chain            0bc39fe Fix CI and add PG17 to the test matrix

[AndrewJackson2020]

Went ahead and tested the upgrade paths.

create chained upgraded extension

PGOPTIONS='--search_path=test' psql -h localhost -d postgres -f ./index_advisor--0.1.0.sql -f index_advisor--0.1.0--0.1.1.sql -f index_advisor--0.1.1--0.1.2.sql -f ./index_advisor--0.1.2--0.2.0.sql

create snapshot upgraded extension

PGOPTIONS='--search_path=test' psql -h localhost -d postgres -f ./index_advisor--0.2.0.sql

did not realize when I initiated the ask that this extension consists of one function. a bit overkill imo as extensions are designed to package and version objects that belong together but that is okay.

\df test.*
                                                                                             List of functions
 Schema |     Name      |                                                                  Result data type                                                                   | Argument data types | Type
--------+---------------+-----------------------------------------------------------------------------------------------------------------------------------------------------+---------------------+------
 test   | index_advisor | TABLE(startup_cost_before jsonb, startup_cost_after jsonb, total_cost_before jsonb, total_cost_after jsonb, index_statements text[], errors text[]) | query text          | func
(1 row)

Either way took the pg_dumps and the end result is that they are the same.

aj@Andrews-MacBook-Pro:~/Repos/mono/external/repos/supabase/index_advisor/v0.2.0/ > diff ./master.sql ../main/chained_upgrade.sql
5c5
< \restrict CO7C6JhNnwyFR93BUbstmEnmOWSIlIZ0JrAiWGuzBJFgIhni92mo1mKVgGWBwIK
---
> \restrict d6Go0Hy35JMAEMkpyOHIPtAxhIcUl3vDK8QxfSzQM7l12K9SpUnFG4W0eHLGgB5
222c222
< \unrestrict CO7C6JhNnwyFR93BUbstmEnmOWSIlIZ0JrAiWGuzBJFgIhni92mo1mKVgGWBwIK
---
> \unrestrict d6Go0Hy35JMAEMkpyOHIPtAxhIcUl3vDK8QxfSzQM7l12K9SpUnFG4W0eHLGgB5j

also some diff output between the relevant files

aj@Andrews-MacBook-Pro:~/Repos/mono/external/repos/supabase/index_advisor/v0.2.0/ > diff index_advisor--0.1.0--0.1.1.sql ../main/index_advisor--0.1.0--0.1.1.sql
diff index_advisor--0.1.0.sql ../main/index_advisor--0.1.0.sql
diff index_advisor--0.1.2.sql ../main/index_advisor--0.1.1--0.1.2.sql
diff index_advisor--0.2.0.sql ../main/index_advisor--0.1.2--0.2.0.sql

0a1,2
> drop function index_advisor(text);
>
aj@Andrews-MacBook-Pro:~/Repos/mono/external/repos/supabase/index_advisor/v0.2.0/ >

[AndrewJackson2020]

Most of these comments are nits. Im satisfied that the new migration path is functionally correct. Good to merge.

[utkarash2991]

Good point - 0.2.1 hasn't shipped anywhere yet. Folded the two additional fixes that were in a separate stacked PR (#15, now closed) directly into #14's 0.2.1 script instead of bumping to 0.2.2. One version, ships once.