On 09.10.23 11:20, Alvaro Herrera wrote:
> I tried this out. I agree it's a good change. BTW, this made me
> realize that "unlike" is not a good name: maybe it should be called
> "except".
right
> I would add quotes to the words "like" and "unlike" there. Otherwise,
> these sentences are hard to parse. Also, some commentary on what this
> is about seems warranted: maybe "Check that this test properly defines
> which dumps the output should match on." or similar.
Done.
I also moved the code a bit earlier, before the checks for supported
compression libraries etc., so it runs even if those cause a skip.
> I didn't like using diag(), because automated runs will not alert to any
> problems. Now maybe that's not critical, but I fear that people would
> not notice problems if they are just noise in the output. Let's make
> them test errors. fail() seems good enough: with the lines I quote
> above and omitting the test corrections, I get this, which seems good
> enough:
After researching this a bit more, I think "die" is the convention for
problems in the test definitions themselves. (Otherwise, you're writing
a test about the tests, which would be a bit weird.) The result is
approximately the same.