illumos gate

I agree with all parties concerned here. The `tr' that was integrated
into illumos was an expedient solution that got the system working
but introduced bugs, some of them still unknown. It's an example
of the conflict between availability and quality. We need to strive
for both, but getting the system working should have priority, at
least in the short term.

Is there a way to mark this integration as a temporary fix, so that
it can be replaced by a better quality fix? Until we have a
comprehensive test suite, we won't even know when we have a
correct version of `tr'.

I'm not aware of any such way. I think we can all agree that this was an expedient solution to the problem... and I'm willing to set aside the current implementation if a better one is available.

So far it looks like the current implementation has some bugs that we now are starting to understand, and that might not be the same bugs in the upstream FreeBSD -- so they may have been introduced by my porting effort, or by the underlying library.

Whether the quickest solution to solving those is to diagnose and fix the bugs here, or to switch to a different implementation is not entirely clear to me. I confess that I'd prefer to understand what the bugs are, if only so that we can fix them if they are in the underlying C library. (Switching out the C library is not on the table.)

In that case, I'd recommend a new bug report be opened.
I see that there already is one relating to certain scripts, but
a more general one might be better.

Issue 132 has been opened for the specific regression, and I think we now have a solution. (I screwed it up.)

If there are other issues known, then please do file a bug.

We need a regression test for this stuff.