Piers Cawley ([EMAIL PROTECTED]) wrote:
> Adam Spiers <[EMAIL PROTECTED]> writes:
> > Adam Spiers ([EMAIL PROTECTED]) wrote:
> >> Think I just found some bad brokenness, although it could be just me
> >> getting hopelessly confused as my brain gets trampled into mush by a
> >> quagmire of exceptions and assertions. I have a testcase:
> >>
> >> sub test_this_should_fail {
> >> my $self = shift;
> >> $self->assert(sub { $_[0] eq $_[1] or die "$_[0] ne $_[1]" }, 'a', 'b');
> >> }
> >>
> >> which does not fail. In fact it behaves identically to
> >>
> >> sub test_this_should_fail {
> >> my $self = shift;
> >> $self->assert(sub { $_[0] eq $_[1] or die "$_[0] ne $_[1]" }, 'a', 'a');
> >> }
> >>
> >> Can anyone reproduce with latest CVS?
> >
> > Ignore that. The problem is that the die() from within a
> > Assertion::CodeRef produces an Error::Simple, which gets turned into a
> > Test::Unit::Failure too far up the stack, so if
> > test_this_should_fail() is in AssertTest.pm, it won't work with
> > check_failures(). Argh. What to do?
>
> Hmm... I'm thinking.
All I've come up with so far is to change check_failures() to
recognise with Error::Simple and treat it as a Test::Unit::Failure.
But AFAICS T::U::Assertion::CodeRef is the only place which seems to
consider a die() as a Test::Unit::Failure rather than a
Test::Unit::Error, and I don't understand how this inconsistency can
work, since a die from inside a T::U::Assertion::CodeRef will be
indistinguishable from (say) a syntax error inside one. I only just
figured out the die() -> Error::Simple -> Test::Unit::Error stuff
yesterday, so I may be misunderstanding things, in which case, please
let me know! If I'm right however, some possible solutions occur to
me:
- Change T::U::Assertion::CodeRef::do_assertion to rethrow die()
exceptions as Test::Unit::Failures. Almost ideal, except for the
serious flaw that syntax errors in the coderef turn up as
failures.
- Change the semantics of T::U::Assertion::CodeRefs so that on
failure they return($failure_msg) rather than die($failure_msg),
and on success they return zero (a la shell)? That would be very
easy to implement, although confusing in that it's the exact
opposite of Perl's notions of true/false.
- Change the semantics of T::U::Assertion::CodeRefs so that on
failure they throw a Test::Unit::Failure rather than a die(). In
that case, it would be better to drop the current requirement that
a coderef has to return true to indicate success, since it would
no longer be necessary.
The third option is my preference.
_______________________________________________
Perlunit-devel mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/perlunit-devel
