At 6:04 PM -0700 4/28/01, Gurusamy Sarathy wrote:
>I'm sure the VMS porters have tried hard enough to make the piping stuff
>work right, but allow me to ask some pesky troublesome questions again
>in the hope that there is still some remote remaining avenue left
>unexplored.
Your questions are better than my answers, but I'll try to describe
where we are and why. Just so we're clear on the relative importance
of this, there are no tests in the current bleadperl that manifest
the problem, but it does come up annoyingly often whenever new tests
are written or new modules with their tests come into the core.
>Is it possible to use some other IPC mechanism that doesn't mess
>with the data itself? Is there something like a Unix domain socket
>or named pipe available on VMS?
There are gizmos called global sections that might work. Or as you
suggest elsewhere, it could possibly be done with temp files (though
I haven't thought about how). Compaq has committed to implementing
named pipes in the kernel in some future version of VMS (I'd guess
it'd be a year or two away, but quite a bit longer before we could
depend on their being available in the majority of installed
systems). The real problem with an alternative approach is that it
implies a complete reimplementation of Chuck Lane's piping code. I
don't know of anyone with both the tuits and the talent to take on
such a project, and for most purposes the current implementation
works very well.
<suggested workarounds snipped>
We really haven't done enough analysis of exactly what goes wrong to
justify carving up pp_print, PerlIO_write, etc. I'll try to nail it
down better and post findings on vmsperl, but at the moment we don't
even have a simple one-or-two-line reproducer.
>I think it would be a shame to change all of the tests to fit the
>VMS straitjacket.
We're not asking anyone to change anything. Nicholas Clark asked how
to avoid the problem, so I told him, but if folks want to leave patching
the tests to us, that's fine. Remember that these patches are rare,
tiny, and indistinguishable from stylistic quirks. I hope as much as
anyone that we won't always have to do them, but until someone finds
time to do something more drastic about the underlying problem,
patching the tests seems the best approach.
--
________________________________________
"... getting out of a sonnet is much more
difficult than getting in."
Brad Leithauser
