Front page | perl.vmsperl |
Postings from October 2002
At 11:06 AM +0100 9/28/02, Steve Wakelin wrote:
>I have an Alphaserver 400 /166 running
>
>o OpenVMS V7.3-1
>o Compaq C V6.2-008 on OpenVMS Alpha V7.3-1
>o %MMK-I-IDENT, this is the MadGoat Make Utility V3.9-3
>
>Perl
>
>o 5.8.0
>o 5.6.1 easy install
>o 5.7.0
>
>when attempting to compile the process always appears to go into a loop
>within MMK when entering the @make_ext
I've been able to reproduce this now with Perl 5.6.1, OpenVMS AXP
v7.3-1, C 6.5, and MMK 2.9-2. The new thing in this configuration is
VMS 7.3-1 with the XFC cache enabled on an 0DS-5 system disk with
hard links enabled. MMS does not exhibit the problem, so there
appears to be something about MMK and VMS 7.3-1 not liking each
other. It consumed 3 hours of cpu time in about 3 hours and 10
minutes of elapsed time and did about 850 million direct I/Os (though
with little apparent disk activity) on my DPW 500au.
I've been poking around with linker maps, SHOW PROCESS/CONTINUOUS,
ANALYZE/SYSTEM, and the output of MMK/LOG and so far haven't gotten
much more precise than to be able to say MMK is chasing its own tail.
I haven't gotten further partly because I've forgotten how to get
from an address where something is happening to a line of source code
where something is happening. The PC values where it gets stuck are
in P2 space, which I think means a system service or AST delivery.
Peeling back the stack by issuing SHOW CALL/NEXT within the SDA
(while logging the output) gives me
$ search sda.log "return address"
Return address on stack = FFFFFFFF.800B6468 EXE$CMODKRNL_C+00198
Return address on stack = 00000000.0004D090 ESIS$DEFINE_PPI+00CD0
Return address on stack = FFFFFFFF.8014D0D0 EXE$AST_RETURN
Return address on stack = 00000000.7AE57230
Return address on stack = 00000000.00043FC0 CTF$SRV$INIT+00A80
Return address on stack = 00000000.00041714 MMK+41714
Return address on stack = 00000000.00040068 UCB$M_SUPMVMSG+00068
Return address on stack = 00000000.7AF7FDD8
Return address on stack = 00000000.7AF7FC18
which I think is telling me that if I can find out what's going on at
41714 bytes (in hex) into the $CODE psect within MMK then I'll have a clue.
Until then, if anyone would like to remind me how this is done or
have a crack at it, please do. In the meantime I think the
workaround is to use MMS or build on 7.3 or earlier.
--
________________________________________
Craig A. Berry
mailto:craigberry@mac.com
"... getting out of a sonnet is much more
difficult than getting in."
Brad Leithauser