+#! /bin/bash
# machten.sh
-# This is for MachTen 4.0.3. It might work on other versions and variants too.
+# This is for MachTen 4.1.4. It might work on other versions and variants
+# too. If it doesn't, tell me, and I'll try to fix it -- domo@computer.org
#
# Users of earlier MachTen versions might need a fixed tr from ftp.tenon.com.
# This should be described in the MachTen release notes.
#
# MachTen 2.x has its own hint file.
#
-# This file has been put together by Andy Dougherty
-# <doughera@lafcol.lafayette.edu> based on comments from lots of
+# The original version of this file was put together by Andy Dougherty
+# <doughera@lafayette.edu> based on comments from lots of
# folks, especially
# Mark Pease <peasem@primenet.com>
# Martijn Koster <m.koster@webcrawler.com>
# Richard Yeh <rcyeh@cco.caltech.edu>
#
+# Deny system's false claims to support mmap() and munmap(); note
+# also that Sys V IPC (re)disabled by jhi due to continuing inadequacy
+# -- Dominic Dunlop <domo@computer.org> 001111
+# Remove dynamic loading libraries from search; enable SysV IPC with
+# MachTen 4.1.4 and above; define SYSTEM_ALIGN_BYTES for old MT versions
+# -- Dominic Dunlop <domo@computer.org> 000224
+# Disable shadow password file access: MT 4.1.1 has necessary library
+# functions, but not header file (or documentation)
+# -- Dominic Dunlop <domo@computer.org> 990804
+# For now, explicitly disable dynamic loading -- MT 4.1.1 has it,
+# but these hints do not yet support it.
+# Define NOTEDEF_MACHTEN to undo gratuitous Tenon hack to signal.h.
+# -- Dominic Dunlop <domo@computer.org> 9800802
+# Completely disable SysV IPC pending more complete support from Tenon
+# -- Dominic Dunlop <domo@computer.org> 980712
+# Use vfork and perl's malloc by default
+# -- Dominic Dunlop <domo@computer.org> 980630
# Raise perl's stack size again; cut down reg_infty; document
# -- Dominic Dunlop <domo@computer.org> 980619
# Use of semctl() can crash system: disable -- Dominic Dunlop 980506
#
# Comments, questions, and improvements welcome!
#
-# MachTen 4.X does support dynamic loading, but perl doesn't
+# MachTen 4.1.1's support for shadow password file access is incomplete:
+# disable its use completely.
+d_getspnam=${d_getspnam:-undef}
+
+# MachTen 4.1.1 does support dynamic loading, but perl doesn't
# know how to use it yet.
+usedl=${usedl:-undef}
+
+# MachTen 4.1.1 may have an unhelpful hack in /usr/include/signal.h.
+# Undo it if so.
+if grep NOTDEF_MACHTEN /usr/include/signal.h > /dev/null
+then
+ ccflags="$ccflags -DNOTDEF_MACHTEN"
+fi
# Power MachTen is a real memory system and its standard malloc
# has been optimized for this. Using this malloc instead of Perl's
-# malloc may result in significant memory savings.
-usemymalloc='false'
+# malloc may result in significant memory savings. In particular,
+# unlike most UNIX memory allocation subsystems, MachTen's free()
+# really does return unneeded process data memory to the system.
+# However, MachTen's malloc() is woefully slow -- maybe 100 times
+# slower than perl's own, so perl's own is usually the better
+# choice. In order to use perl's malloc(), the sbrk() system call
+# must be simulated using MachTen's malloc(). See malloc.c for
+# precise details of how this is achieved. Recent improvements
+# to perl's malloc() currently crash MachTen, and so are disabled
+# by -DPLAIN_MALLOC and -DNO_FANCY_MALLOC.
+usemymalloc=${usemymalloc:-y}
+
+# Older versions of MachTen malloc() data on a two-byte boundary, which
+# works, but slows down operations on long, float and double data.
+# Perl's malloc() can compensate if SYSTEM_ALLOC_ALIGNMENT is suitably
+# defined.
+if expr "$osvers" \< "4.1" >/dev/null
+then
+system_alloc_alignment=" -DSYSTEM_ALLOC_ALIGNMENT=2"
+fi
+# Do not wrap the following long line
+malloc_cflags='ccflags="$ccflags -DPLAIN_MALLOC -DNO_FANCY_MALLOC -DUSE_PERL_SBRK$system_alloc_alignment"'
+
+# When MachTen does a fork(), it immediately copies the whole of
+# the parent process' data space for the child. This can be
+# expensive. Using vfork() where appropriate avoids this cost.
+d_vfork=${d_vfork:-define}
+
+# Specify a high level of optimization (-O3 wouldn't do much more)
+optimize=${optimize:--O2 -fomit-frame-pointer}
# Make symbol table listings les voluminous
nmopts=-gp
X=`expr $X / 2`
stack_size=`expr $stack_size \* 2`
done
+ X=`expr $stack_size \* 1024`
fi
-X=`expr $stack_size \* 1024`
ldflags="$ldflags -Xlstack=$X"
ccflags="$ccflags -DREG_INFTY=$reg_infty"
# friends. Use setjmp and friends instead.
expr "$osvers" \< "4.0.3" > /dev/null && d_sigsetjmp='undef'
-# semctl(.., .., IPC_STATUS, ..) hangs system: say we don't have semctl()
-d_semctl='undef'
+# System V IPC before MachTen 4.1.4 is incomplete (missing msg function
+# prototypes, no ftok()), buggy (semctl(.., .., IPC_STATUS, ..) hangs
+# system), and undocumented. Claim it's not there at all before 4.1.4.
+if expr "$osvers" \< "4.1.4" >/dev/null
+then
+d_msg=${d_msg:-undef}
+d_sem=${d_sem:-undef}
+d_shm=${d_shm:-undef}
+fi
+
+
+# As of MachTen 4.1.4 the msg* and shm* are in libc but unimplemented
+# (an attempt to use them causes a runtime error)
+# XXX Configure probe for really functional msg*() is needed XXX
+# XXX Configure probe for really functional shm*() is needed XXX
+if test "$d_msg" = ""; then
+ d_msgget=${d_msgget:-undef}
+ d_msgctl=${d_msgctl:-undef}
+ d_msgsnd=${d_msgsnd:-undef}
+ d_msgrcv=${d_msgrcv:-undef}
+ case "$d_msgget$d_msgsnd$d_msgctl$d_msgrcv" in
+ *"undef"*) d_msg="$undef" ;;
+ esac
+fi
+if test "$d_shm" = ""; then
+ d_shmat=${d_shmat:-undef}
+ d_shmdt=${d_shmdt:-undef}
+ d_shmget=${d_shmget:-undef}
+ d_shmctl=${d_shmctl:-undef}
+ case "$d_shmat$d_shmctl$d_shmdt$d_shmget" in
+ *"undef"*) d_shm="$undef" ;;
+ esac
+fi
+
+# MachTen has stubs for mmap and munmap(), but they just result in the
+# caller being killed on the grounds of "Bad system call"
+d_mmap=${d_mmap:-undef}
+d_munmap=${d_munmap:-undef}
# Get rid of some extra libs which it takes Configure a tediously
-# long time never to find on MachTen
+# long time never to find on MachTen, or which break perl
set `echo X "$libswanted "|sed -e 's/ net / /' -e 's/ socket / /' \
-e 's/ inet / /' -e 's/ nsl / /' -e 's/ nm / /' -e 's/ malloc / /' \
-e 's/ ld / /' -e 's/ sun / /' -e 's/ posix / /' \
- -e 's/ cposix / /' -e 's/ crypt / /' \
+ -e 's/ cposix / /' -e 's/ crypt / /' -e 's/ dl / /' -e 's/ dld / /' \
-e 's/ ucb / /' -e 's/ bsd / /' -e 's/ BSD / /' -e 's/ PW / /'`
shift
libswanted="$*"
During Configure, you may see the message
*** WHOA THERE!!! ***
- The recommended value for \$d_semctl on this machine was "undef"!
+ The recommended value for \$d_msg on this machine was "undef"!
Keep the recommended value? [y]
-Select the default answer: semctl() is buggy, and perl should be built
-without it.
+as well as similar messages concerning \$d_sem and \$d_shm. Select the
+default answers: MachTen 4.1 appears to provide System V IPC support,
+but it is incomplete and buggy: perl should be built without it.
+Similar considerations apply to memory mapping of files, controlled
+by \$d_mmap and \$d_munmap.
+
+Similarly, when you see
+
+*** WHOA THERE!!! ***
+ The recommended value for \$d_vfork on this machine was "define"!
+ Keep the recommended value? [y]
+
+select the default answer: vfork() works, and avoids expensive data
+copying.
+
+You may also see "WHOA THERE!!!" messages concerning \$d_getspnam.
+Select the default answer: MachTen's support for shadow password
+file access is incomplete, and should not be used.
At the end of Configure, you will see a harmless message
Hmm...You had some extra variables I don't know about...I'll try to keep 'em.
Propagating recommended variable dont_use_nlink
Propagating recommended variable nmopts
+ Propagating recommended variable malloc_cflags...
Propagating recommended variable reg_infty
+ Propagating recommended variable system_alloc_alignment
Read the File::Find documentation for more information about dont_use_nlink
Your perl will be built with a stack size of ${stack_size}k and a regular
https://perl5.git.perl.org / perl5.git / blobdiff