Message-ID: <6ccbe513-a815-4d2c-a746-abe5de2da86b@iki.fi>
Date: Mon, 9 Feb 2026 12:02:10 +0200
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: Fix a bug in extension_file_exists()
To: Chao Li <li.evan.chao@gmail.com>,
 Postgres hackers <pgsql-hackers@lists.postgresql.org>
References: <362EA9B3-589B-475A-A16E-F10C30426E28@gmail.com>
Content-Language: en-US
From: Heikki Linnakangas <hlinnaka@iki.fi>
In-Reply-To: <362EA9B3-589B-475A-A16E-F10C30426E28@gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Archived-At: <https://www.postgresql.org/message-id/6ccbe513-a815-4d2c-a746-abe5de2da86b%40iki.fi>
Precedence: bulk

On 09/02/2026 11:21, Chao Li wrote:
> I just noticed a bug in extension_file_exists():
> 
> ```
> bool
> extension_file_exists(const char *extensionName)
> {
> 	bool		result = false;
> 	List	   *locations;
> 	DIR		   *dir;
> 	struct dirent *de;
> 
> 	locations = get_extension_control_directories();
> 
> 	foreach_ptr(char, location, locations) // <== Here type char is wrong
> 	{
> 		dir = AllocateDir(location);
> ```
> 
> get_extension_control_directories() returns a list of ExtensionLocation, but the loop iterates it as if it contained char *, which is incorrect. As a result, AllocateDir() and ReadDir() are called with the wrong type.
> 
> This bug is only triggered on an error path, when PostgreSQL is deciding whether to emit a hint. For example:
> ```
> evantest=# create function f() returns int LANGUAGE plpython3u as $$return 1$$;
> ERROR:  language "plpython3u" does not exist
> ```
> No hint is printed.
> 
> With this patch applied:
> ```
> evantest=# create function f() returns int LANGUAGE plpython3u as $$return 1$$;
> ERROR:  language "plpython3u" does not exist
> HINT:  Use CREATE EXTENSION to load the language into the database.
> ```
> So the hint is shown as intended.
> 
> Attached is a patch fixing the iteration to use ExtensionLocation and location->loc consistently.

Yep, good catch. This went wrong in commit f3c9e341cd, which changed the 
type of objects in the list from "char *" to "ExtensionLocation *". So 
this is master only, stable branches are not affected.

I will push the fix shortly, thanks!

- Heikki