Greetings,
* Michael Paquier (michael@paquier.xyz) wrote:
> On Thu, Mar 05, 2020 at 07:32:59PM -0500, David Steele wrote:
> > FWIW, we use static values in our unit tests (which are written in C), then
> > test against packaged versions of Postgres for integration tests.
> >
> > When I saw the commit I pulled it in so I could remove instructions for the
> > manual step to add the cast. So in this case the issue was apparent really
> > quickly. Normally we only pull in new code from PostgreSQL once a year.
> >
> > We think our unit tests against static values may have endianess issues but
> > we have not verified that one way or the other. Here's what they look like:
> >
> >
https://github.com/pgbackrest/pgbackrest/blob/e55443c890181ea63a350275447885331c8254e4/test/src/module/postgres/interfaceTest.c#L182
>
> By doing so, the tests still fail if the page size is something else
> than 8k, no?
Of course, but we check that from pg_controldata before we actually
start doing anything with the cluster.
Even a test which at least validated that the checksum code was
returning the right value when page size is the (rather common...) 8K
would be better than a change like this not breaking any tests..
Thanks,
Stephen