On Thu, Nov 06, 2025 at 12:45:30PM -0500, Andres Freund wrote:
> Could we add a testcase that actually exercises at least some of the
> codepaths? We presumably wouldn't want to actually write that much data, but
> it shouldn't be hard to write portable code to create a file with holes...
With something that relies on a pg_pwrite() and pg_pread(), that does
not sound like an issue to me.
FWIW, I have wanted a test module that does FS-level operations for
some time. Here, we could just have thin wrappers of the write and
read calls and a give way for the tests to pass directly arguments to
them via a SQL function call. That would be easier to extend
depending on what comes next. Not sure that this is absolutely
mandatory for the sake of the proposal, though, but long-term that's
something we should do more to stress the portability of the code.
--
Michael
