On Tue, Apr 8, 2025 at 10:36 AM Jacob Champion
<jacob.champion@enterprisedb.com> wrote:
> Yeah, but it's one of those things that feels like it must have been
> solved by the others in the space. Once it's installed, the concern
> goes away (unless you demand absolute relocatability without
> recompilation). I'll take a look at how libkrb/libmagick do their
> testing.
Perhaps unsurprisingly, they inject different lookup paths via
envvars. We could do the same (I have FUD about the security
characteristics)...
> If it somehow turns out to be impossible, one option might be to shove
> a more detailed ABI identifier into the name.
...but I wonder if I can invert the dependency on
libpq_append_conn_error entirely, and remove that part of the ABI
surface, then revisit the discussion on `-<major>.so` vs
`-<major>-<minor>.so`.
--Jacob
