Re: [cp-patches] FYI: Remove (bogus) asserts in java-net.

[Tom Tromey]

>>>>> "Mark" == Mark Wielaard <address@hidden> writes:

Mark> I find the example a bit contrived. You will get a crash anyway in f2()
Mark> since you trashed env and are using it.

Yes... let's not talk about asserts in the abstract.  I don't think
that sheds much light.

>> >   assert((*env)->GetArrayLength(env, receivedFromAddress) > 4);

Mark> The asserts failed because they were testing the wrong condition (they
Mark> test for > instead of ==).

Oops.

Mark> This code is so tightly bound to the private
Mark> VM method that I don't feel argument assert checking buys us anything.

If the code in question lives in a different file, then I think that
having the assert does add something -- it ensures that a change in
one place has a chance of being caught in the other.  This can be very
useful.  In this case, fixing the assertion to be correct certainly
wouldn't hurt anything, and would provide some mild protection against
changes.

Another question to ask is whether an assert is the appropriate
self-checking mechanism at a given point.  If it is purely
classpath-internal stuff, then this is probably ok.  That is
especially true if our test suite actually runs this code path -- that
way we can be reasonably sure that a Classpath release won't trigger
the assertion.

In some other situations a Java exception is probably more
appropriate.

Tom