cpython/Modules/make_aix_so

123 lines
6.5 KiB
Bash
Executable File

#!/bin/ksh
#################################
# AIX shared library helper #
#################################
# ========================================================================
# FILENAME: make_aix_so
# MODULE FOR: standalone executable
# PLATFORM: AIX (specific)
# DESCRIPTION: Creates a shareable .o from a pre-compiled (unshared)
# .o file
# ARGUMENTS: Same as for "ld". The -bM, -bE, -bI, -H, -T, and -lc
# arguments of "ld" will be supplied by this script.
# NOTES: 1. Currently specific to the building of Python
# interpreter shared objects, in that the entry
# point name is hardcoded based on the object file
# name (the "mathmodule.o" file will expect an
# entry point of "initmath"). This could be remedied
# by the support (or simple expectation) of a "-e"
# argument.
# 2. The resulting shared object file is left in the
# current directory with the extension .so. It may
# need to be changed to have a .o extension before
# it is usable. (At least, Python expects it to
# have the .o extension, but this is simply because
# python wants it that way -- it COULD probably be
# called anything at all).
# HISTORY: Manus Hand (mhand@csn.net) -- Initial code -- 6/24/96
# ========================================================================
# ========================================================================
# SET UP VARIABLES FOR USE IN THIS SCRIPT
# ------------------------------------------------------------------------
# Note that the setting of "entry" is Python-build specific. This script
# is not general-purpose for that reason (although support for a "-e"
# argument to it could be added, making it usable for any AIX application)
# ========================================================================
objfile=$1
shift
filename=`echo $objfile | sed -e "s:.*/\([^/]*\)$:\1:" -e "s/\..*$//"`
entry=init`echo $filename | sed "s/module.*//"`
ldargs="-e$entry -bE:$filename.exp -bM:SRE -T512 -H512 -lc $objfile $*"
tmpfile=.py_$$
# ======================================================================
# EXPORT LIST GENERATION
# ----------------------------------------------------------------------
# For the Python modules, this COULD be much simpler, since we know the
# only exported variable is ".$entry" ("entry" was assigned just above).
# However, the approach used here for export file generation is more
# generic and will support all .o's, not just properly formatted Python-
# importable modules. Here is the rule: any "extern" symbol name which
# appears in the # output of "nm" which IS resolved (i.e., which does
# NOT have an address of zero) should go into the export list. Read
# each line from a temp file containing these symbols. If it begins
# with a dot, then add it to the list being built. If it does not, then
# see if the same symbol, with the dot prepended, also appears in the
# list. If so, DON'T include the current symbol (the one without the
# prepended dot).
# ======================================================================
exec 3>&1 1>$filename.exp
echo "#!$objfile"
nm $objfile | grep "|extern|" | grep -v " 0|extern|" | cut -f1 -d"|" > $tmpfile
while read line ; do
echo "$line" | cut -c1 | read prefix
if [ "$prefix" = "." ]; then
echo "$line"
else
grep "^\.$line" $tmpfile > /dev/null
if [ $? != 0 ]; then
echo "$line" ; fi ; fi ; done < $tmpfile
rm $tmpfile
# ===============================================================
# IMPORT LIST AND SHARED OBJECT FILE GENERATION
# ---------------------------------------------------------------
# Send all output to the to-be-built import file, starting it off
# with the required "#!" line (which tells it in which loaded
# binary these symbols are to be found at runtime). Then attempt
# to ld the object using only the export list built above, and
# hide the stderr output from "ld". If the ld fails with status
# code 8 (and in the case of the Python modules, it always does,
# since each need some symbols from the statically linked portion
# of the interpreter), this is because an import list should be
# given containing the symbols which are unresolved. The symbols
# will have been sent to stdout as a side-effect of the failed ld
# command, so by redirecting the stdout output, they will have
# magically been put into the import file being built. Then we
# simply call ld again with both the import and export lists.
# ===============================================================
exec 1>$filename.imp
echo "#!python"
ld $ldargs 2>/dev/null
status=$?
exec 1>&3
# ================================================================
# GUIDO: If you want to separate the generation of the import and
# export lists from the creation of the .so file, here's where the
# break should be made -- in my mail I mentioned that some of this
# script belongs in the pre-static link stage of the make and some
# belongs after it. As I said, here is the dividing line. Now,
# of course, if there is a module which needs NO statically linked
# symbols -- but then again, there can't be, because they all need
# initmodule4() -- the "ld" which is ABOVE this line may actually
# have succeeded, so the "if" below will fail, but of course,
# if you separate the script at this point, you won't care about
# such things.
# ================================================================
if [ $status = 8 ] ; then
ld $ldargs $filename.imp ; fi
# ======================================================================
# GUIDO: Remember that at this point, the files (assuming you leave the
# arguments to LDSHARED totally unchanged) are still named with a .so
# extension. However, Python looks for them with a .o extension. You
# can either change this in the core code (#ifdef _AIX) so that it looks
# for an .so or you can do what I did, which is rename them to .o's when
# they get mv'ed by the sharedinstall make rule. (Actually, I did it by
# hand, but you'd do it in sharedinstall.
# =======================================================================