[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Sat, 29 Mar 2003 16:44:36 +0100
Gnus/5.090016 (Oort Gnus v0.16) Emacs/21.2 (gnu/linux)
| Akim Demaille wrote:
| >I am firmly against this patch, and ask for it to be reversed. You
| >will first have to explain it, and teach us why we'd want such a code
| >bloat. The very first problem with this patch is its ChangeLog.
| It allows scripts to appear in autotest sources, making an autotest
| script behave much more like configure.in.
It was designed this way. Please, revert this patch.
| It allows the CVS test suite to set some variables, for instance
| $CVSROOT, for the tests to share.
You just demonstrated why we do not want this to happen: there should
be no dependency whatsoever with previous states. If you want to set
CVSROOT in many places, then do introduce AT_CHECK_CVS that sets it.
| In conjunction with my option
| parsing patch, this can be switched depending on whether the user
| wanted to test the client/server or local mode of CVS.
| It will also allow some constructs such as:
| for remote in false :; do
| for server in cvs-1.10 cvs-1.11 cvs-1.11.5; do
| Where tests.at contains AT_SETUP...AT_CHECK...AT_CLEANUP
If you want to run this, you must do it in M4, not in sh.
| Tests defined in tests.at have access to all the environment variables
| defined previously, including the `remote' and `server' variables from
| the above examples.
They must not. You must reset the environment for each test.
| This patch increases the size of the test suite slightly, but disk is
| cheap and portability is priceless.
My concern is determinism of the test suite. My concern is the full
modularity of the tests: they do not depend upon the past in the test
Please, revert this patch, and let's address your issue in proper Autotest.