Matthias Pfaller wrote:
>Ian Dall wrote:
>> This is due to a bug in libc memmove and bcopy.
>> I can confirm that when libc is fixed, perl works on the pc532.
>> I understand, (but haven't tried) that memmove and bcopy have been
>> fixed in current.
>
>Oh sorry, I messed up again :-( After all you told me where you stumbled
>over the bug, but the PR didn't trigger my memory. Yes, thank's to Ian's
>diagnosis I was able to fix the bug. The -current libc shouldn't have
>the problem Ian describes. I have appended what I changed in -current.
>
>Matthias
>
>RCS file: /cvsroot/src/lib/libc/arch/ns32k/string/memcpy.S,v
>diff -u -r1.4 -r1.6

[Patch deleted except for this line:]

>+	ret	ARGS

I added the patch. (Manually, I was hoping patch would do it for
me, but alas... 8-)  I assume that "ARGS" should be replaced by
"0" as in the other return statements.  (As evidenced by the fact
that it seems to work with "0" and ld.so complains about ARGS not
being defined when built that way. 8-)

I can now confirm that the perl5 package builds fine, but I haven't
installed or tested it yet.  (I wonder if this was causing the
emacs problems I was having as well...?) I am now in the midst of
a "stage2" build of NetBSD-1.3.2/pc532.  The 1.3.2 sources built
just fine from the 1.3.1 binaries, and so far, 1.3.2 is building
OK with the "stage1" binaries.

I suppose, technically, I should bump libc.12.20 to .21, but I
didn't.  I suspect that would cause other large problems.  Perhaps
we should add a last minute note into the upgrade/install docs
stating that a nasty bug was fixed, but the library version number
is unchanged.  (On the other hand, the pc532 community is small
enough that they might all be fully aware of the problem by now.
8-)

I now strongly recommend just closing the PR stating that it's not
the package, but a port-specific library bug.

Jon