[PATCH] Add hint for misspelled relations - Mailing list pgsql-hackers

From Steve Chavez
Subject [PATCH] Add hint for misspelled relations
Date
Msg-id CAGRrpzaBQdPdSRXwHnLnCvs=f1a1abcecgBq5f5s+XPGU7f_zQ@mail.gmail.com
Whole thread Raw
Responses Re: [PATCH] Add hint for misspelled relations
List pgsql-hackers
Hello hackers,

Currently misspelled columns offer a hint but not misspelled relations.

This patch enables that, the result is like:

-- having this table
create table clients (id int); 
-- we misspell the table name 
select * from cliants;
ERROR:  relation "cliants" does not exist
LINE 1: select * from cliants;
HINT:  Perhaps you meant to reference the table "public.clients".

The possible matches are searched in pg_class for the schemas present in search_path (or if the relation is qualified, it only searches matches in that schema). The logic reuses the `varstr_levenshtein_less_equal` function similar to how the column matching is done. 

If there's a tie in the fuzzy match, it's solved by preferring the schema that appears first on the search path. If that fails, then the lexicographic order is used to break the tie.

One problem is that scanning all pg_class entries can get expensive on big catalogs, so the number of searches is capped by MAX_REL_HINT_CANDIDATES. I've set this to 4096 arbitrarily, any guidance on what would be a good number is appreciated. Personally I've seen  a catalog that contains 125K tables, with mostly auto generated names. For these cases I don't think the hint helps that much anyway, so it seemed fine to bail here.

The changes are split into two commits, one refactoring some reusable functions for easier review and another one implementing the relation hint.

Any feedback is welcomed.

Best regards,
Steve Chavez

[1]:
Attachment

pgsql-hackers by date:

Previous
From: Sutou Kouhei
Date:
Subject: Re: Make COPY format extendable: Extract COPY TO format implementations
Next
From: Michael Paquier
Date:
Subject: Re: [Patch] Windows relation extension failure at 2GB and 4GB