Re: make installcheck-world in a clean environment

Your proposed fix involves inventing something called
USE_INSTALLED_ASSETS, but we don't need that anywhere else, so why do
we need it for ECPG?

We need that not only for ECPG, but for libpostgres, libpgport, and libpq too, if we want to check the installed ones (instead of recompiled).
Imagine that we will recompile psql (and use it) to perform installcheck for the server instance. It will work but we'll miss an opportunity to check whether the installed psql is working. I believe, it's the same with ecpg (as now, we can even remove installed ecpg and have all the tests passed).

I am not opposed to fixing this, but one might also ask whether it's a
good use of time, since the workaround is straightforward.

I'm not seeing a workaround to perform more complete installcheck without modifying makefiles. So for me the question is whether the increasing the testing surface justifies this use of time.
Best regards,

------
Alexander Lakhin
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company