Re: Another ash braindamage (testsuite only)

[Pavel Roskin]

Hello, Akim!

> | The problem is that the code immediately after that relies on $?. In fact,
> | $? is normally (in bash) reset to 0 by executing $at_traceon, which is
> | normally empty because at_traceon is not set.
> 
> What code are you referring to?  Who is depending upon $? after
> running this?  Could you develop?

AT_CHECK (defined in tests/atgeneral.m4) uses to have the following code:

$at_failed && exit 1

at_failed is set initially to "false" and then to ":" if any of diffs
(for stdout or stderr) fails. Now, let's see what happens if both diffs
succeed.

at_failed remains "false", thus we have

false && exit 1

In this case we don't exit, but $? is set to 1 by "false". Now if the
subshell terminates at this point, $? remains 1, which means that
AT_CLEANUP would report an error.

But $at_traceon in the end of AT_CHECK changes $? to 0. Note that
at_traceon is either undefined or defined to "set +vx"

When at_traceon is not defined $? is still cleared in most shells but not
in ash-0.2.

> This should be described in the documentation, along with all the
> limitations you've already discovered :)

Ok, I'll try to find time for that.

Regards,
Pavel Roskin