On Sat, 11 May 2002, Martti Kuparinen wrote:

> > I think I know why you want to do it this way, but I don't like that way.
>
> Me either, but I wouldn't like to change all tests to match our layout, it
> would be easier to use the tests as they are in the original distribution.
>
> I'll re-consider this today...

I've considered this again and came to the following conclusions:

* Regression tests are performed in /usr/src/regress

* /usr/src/dist/ipf should contain unmodified(?) distribution of ipf

* We need to change some tests (to fix errors) but dist/ipf/test should be
  left untouched!?

* It's possible to create a new Makefile in regress/sys/kern/ipf (we have
  already done this in usr.sbin/ipf/ipftest/test) but it's easy to forget
  later as we have previously seen with usr.sbin/ipf/ipftest/test/Makefile.

  The problem here is that original tests are not here, i.e. there's no
  DARRENR branch available. Without that vendor branch we only have a locally
  created file (like in usr.sbin/ipf/ipftest/test). How should we handle
  updates to this file? Get the official file and modify it manually every
  time? I don't like this at all...

* Importing the original tests to regress/sys/kern/ipf is good because
  it's easy to modify the main trunk with our local fixes and it's still
  easy to merge new versions to main trunk when new versions of ipf are
  imported (i.e. the same thing we are doing today with sys/netinet etc).

Based on these, my proposal is this:

* There will be 4 different locations with DARRENR vendor branch (dist/ipf,
  sys/netinet, sys/lkm/netinet/if_ipl and regress/sys/kern/ipf) instead of
  the 3 we have today.

* Remove dist/ipf/test

Does this sound reasonable?

Martti

---
Martti Kuparinen <martti.kuparinen@iki.fi>      NetBSD - No media hype
http://www.iki.fi/kuparine/                     http://www.netbsd.org/