#!/usr/bin/env python

import os, sys

if len(sys.argv) != 3:
    print "%s: <index> <pkgdir>" % sys.argv[0]
    sys.exit()

indexfile = sys.argv[1]
pkgdir = sys.argv[2]

if not pkgdir.endswith("/All"):
    pkgdir = pkgdir + "/All"

packages = [pkg for (pkg, ext) in map(os.path.splitext, os.listdir(pkgdir)) if ext == ".tbz"]

index=[]
pkgs=[]
for i in file(indexfile):
    out = i.rstrip().split("|")
    out[7] = out[7].split(" ") # build dep
    out[8] = out[8].split(" ") # run dep
    index.append(out)

    # Keep track of all the packages we have seen in the index.  In
    # principle there is no need to track the build/run deps since
    # they will also be listed in field 0.  We could add a sanity
    # check for this.
    pkgs.append(out[0])
    pkgs.extend(out[7])
    pkgs.extend(out[8])

used=set(pkgs)
notfound=used.difference(set(packages))

# Write out the new index, stripping out the entries for missing
# packages as well as dependencies from existing packages on the
# missing ones.
#
# This is slightly dubious since it will intentionally list packages
# that are present but missing dependencies on non-redistributable
# things like jdk that were successfully built but removed already, so
# the dependency lists will not be complete.  It matches the old
# chopindex.sh behaviour though.
#
# I think it would be better to just prune those incomplete packages
# from the INDEX altogether, but I don't know if anyone is relying on
# this historical behaviour.

for data in index:
    if data[0] not in notfound:
        print "%s|%s|%s|%s" % ("|".join(data[:7]),
                               " ".join([j for j in data[7] if j not in notfound]),
                               " ".join([j for j in data[8] if j not in notfound]),
                               "|".join(data[9:]))