Re: Potential problem in commit f777d773878 and 4f7f7b03758
Dilip Kumar <dilipbalaut@gmail.com>
From: Dilip Kumar <dilipbalaut@gmail.com>
To: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Cc: Peter Eisentraut <peter@eisentraut.org>
Date: 2025-08-22T11:02:34Z
Lists: pgsql-hackers
Attachments
- poc_fix.patch (application/octet-stream) patch
On Fri, Aug 22, 2025 at 4:04 PM Dilip Kumar <dilipbalaut@gmail.com> wrote: > > Simple test to reproduce: > > Change the module path for any extension as below, like I have done > for amcheck and then copy the .so file at $libdir/test/ folder. > module_pathname = '$libdir/test/amcheck' > > While creating the extension it fail to load the file because, after > commit f777d773878 we remove the $libdir from the path[1] and later in > expand_dynamic_library_name() since there is a '/' in remaining path > we will call full = substitute_path_macro(name, "$libdir", > pkglib_path); to generate the full path by substituting the "$libdir" > macro[2]. But we have already removed $libdir from the filename, so > there will be no substitution and it will just search the > 'test/amcheck' path, which was not intended. > > IMHO the fix should be that we need to preserve the original name as > well, and in the else case we should pass the original name which is > '$libdir/test/amcheck' in this case so that macro substitution can > work properly? > I just quickly changed this and it's working fine, but I haven't analyzed fully whether it breaks something which was intended by the commit (f777d773878 and 4f7f7b03758) but check-world is passing. -- Regards, Dilip Kumar Google
Commits
-
Fix: Don't strip $libdir from nested module_pathnames
- d07e2d4237cb 18.0 landed
- 990c8db1827c 19 (unreleased) landed
-
Don't strip $libdir from LOAD command
- f777d773878d 18.0 cited