Fw: Re: [nhc-bugs] problems with 1.12

Malcolm Wallace Malcolm.Wallace@cs.york.ac.uk
Wed, 20 Mar 2002 17:32:16 +0000


Hi Feliks,

> > (Forwarded message, original mistakenly sent only to Feliks.)
> Never got the first one.

[ I got a "not-delivered after 24 hours" warning on the original,
  which alerted me to the fact it was addressed only to you. ]

> It's just that I was so impressed with the ease of porting the previous
> two versions that I wondered whether there was anything in your
> "process" that gave this effect and that you might be able to restore. 

Well it is true that at one time we built the compiler regularly
on three different machine architectures (ix86-Linux, mips-IRIX,
sparc-solaris).  We no longer have any mips-IRIX machines, and the
nightly sparc-solaris build is currently on hold due to an annoying
feature of SSH, so there is only one machine/OS combination being
tested daily at the moment.

You are absolutely right that we should try to restore this part of
our "process", since I am sure that its lack is the cause of many of
the small errors you and others have been seeing.

> Since hbc more or less expired, and ghc is excellent, but devilishly
> difficult to port, I would imagine that portability could be _the_
> "selling" point for nhc, now that memory has become so cheap.

We have always sold nhc98 on portability, and would like to continue
to do so.

> The file is there, and the symbols are there, too.
>
> I have a vague recollection that half a year ago (in another context,
> nothing to do with nhc) I encountered a similar problem and that it had
> something to do with assumptions about whether a library file is in
> shareable format...

I have a suggestion - try running `ranlib' on the archive file.
I think the GNU ar does this automatically at creation time, but the
BSD ar probably does not.  If this fixes the problem, then I will
add it to the Makefile.

> I don't even know what "observe" does, so this might be irrelevant.

It implements fast file-access primitives for the hat-observe program,
which is (obviously) one of the Hat trace-browsing tools.  If you
aren't intending to use tracing (why not?! :-) then this part of the
build doesn't matter to you anyway.

Regards,
    Malcolm