Re: Readd use of TAP subtests

=?utf-8?Q?Dagfinn_Ilmari_Manns=C3=A5ker?= <ilmari(at)ilmari(dot)org> writes:
> The only cases where an explicit plan adds value is if you're running
> tests in a loop and care about the number of iterations, or have a
> callback with a test inside that you want to make sure gets called. For
> these, it's better to explicitly assert that the list you're iterating
> over is of the right length, or increment a counter in the loop or
> callback and assert that it has the expected value. This has the added
> benefit of the failure being coming from the relevant place and having a
> helpful description, rather than a plan mismatch at the end which you
> then have to hunt down the cause of.

Yeah. A different way of stating that is that the test count adds
security only if you re-derive its proper value from first principles
every time you modify the test script. I don't know about you guys,
but the only way I've ever adjusted those numbers is to put in whatever
the error message said was right. I don't see how that's adding
anything but make-work; it's certainly not doing much to help verify
the script's control flow.

A question that seems pretty relevant here is: what exactly is the
point of using the subtest feature, if we aren't especially interested
in its effect on the overall test count? I can see that it'd have
value when you wanted to use skip_all to control a subset of a test
run, but I'm not detecting where is the value-added in the cases in
Peter's proposed patch.

regards, tom lane