Re: Avoid certain spurious `testsuite' rebuilds

[Noah Misch]

On Mon, Apr 10, 2006 at 08:01:44PM +0200, Stepan Kasal wrote:
> On Mon, Apr 10, 2006 at 02:41:38PM +0200, Ralf Wildenhues wrote:
> > The trap code touches several files.  Isn't that wrong for them to have
> > new time stamps in the error case?
> 
> Yes, you are right.

If `mktests.sh' might be unportable, the trap helps.  Suppose a user edits one
of lib/autoconf/*.m4 in an effort to fix a bug on his/her platform.  He or she
then runs `make check', and `mktests.sh' stumbles.  If it touches tests/*.at
before exiting with an error, the user can run `make check' again and proceed
with the test suite.  Otherwise, the user can't use 'make check' until he or she
also fixes `mktests.sh' for his/her system.

> Perhaps the following changes should go in right now?
> 1) mktests.sh no longer tolarates bugs
> 2) thus the trap doesn't touch the *.at files

I'm not entirely sure what you mean by (1), but this doesn't seem like the kind
of change to put in at this stage.  What do we gain?

> 3) "trap '' 0" might be more portable

This looks fine, but you wrote `test' instead of `trap' in your ChangeLog entry:

>       * tests/mktests.sh: Use "test '' 0", not "test 0".  Do not touch
>       the files if a problem appears.  If *.at is going to be empty,
>       exit with an error.  Proposed by Ralf Wildenhues.