def itermergesort(sources, key, header, missing, reverse):
# first need to standardise headers of all input tables
# borrow this from itercat - TODO remove code smells
its = [iter(t) for t in sources]
source_flds_lists = [it.next() for it in its]
if header is None:
# determine output fields by gathering all fields found in the sources
outflds = list()
for flds in source_flds_lists:
for f in flds:
if f not in outflds:
# add any new fields as we find them
outflds.append(f)
else:
# predetermined output fields
outflds = header
yield tuple(outflds)
def _standardisedata(it, flds, outflds):
# now construct and yield the data rows
for row in it:
try:
# should be quickest to do this way
yield tuple(row[flds.index(f)] if f in flds else missing
for f in outflds)
except IndexError:
# handle short rows
outrow = [missing] * len(outflds)
for i, f in enumerate(flds):
try:
outrow[outflds.index(f)] = row[i]
except IndexError:
pass # be relaxed about short rows
yield tuple(outrow)
# wrap all iterators to standardise fields
sits = [
_standardisedata(it, flds, outflds)
for flds, it in zip(source_flds_lists, its)
]
# now determine key function
getkey = None
if key is not None:
# convert field selection into field indices
indices = asindices(outflds, key)
# now use field indices to construct a _getkey function
# N.B., this will probably raise an exception on short rows
getkey = operator.itemgetter(*indices)
# OK, do the merge sort
for row in shortlistmergesorted(getkey, reverse, *sits):
yield row
def _mergesorted(key=None, reverse=False, *iterables):
# N.B., I've used heapq for normal merge sort and shortlist merge sort for reverse
# merge sort because I've assumed that heapq.merge is faster and so is preferable
# but it doesn't support reverse sorting so the shortlist merge sort has to
# be used for reverse sorting. Some casual profiling suggests there isn't much
# between the two in terms of speed, but might be worth profiling more carefully
if reverse:
return shortlistmergesorted(key, True, *iterables)
else:
return heapqmergesorted(key, *iterables)
def close(self):
# sort anything remaining in the cache
self.cache.sort(key=self.getkey, reverse=self.reverse)
if self.chunkfiles:
chunkiters = [iterchunk(f) for f in self.chunkfiles]
chunkiters.append(self.cache) # make sure any left in cache are included
for row in shortlistmergesorted(self.getkey, self.reverse, *chunkiters):
self.broadcast(row)
else:
for row in self.cache:
self.broadcast(row)
super(SortConnection, self).close()
def close(self):
# sort anything remaining in the cache
self.cache.sort(key=self.getkey, reverse=self.reverse)
if self.chunkfiles:
chunkiters = [iterchunk(f) for f in self.chunkfiles]
chunkiters.append(self.cache) # make sure any left in cache are included
for row in shortlistmergesorted(self.getkey, self.reverse, *chunkiters):
self.broadcast(row)
else:
for row in self.cache:
self.broadcast(row)
super(SortConnection, self).close()
def itermergesort(sources, key, header, missing, reverse):
# first need to standardise headers of all input tables
# borrow this from itercat - TODO remove code smells
its = [iter(t) for t in sources]
source_flds_lists = [it.next() for it in its]
if header is None:
# determine output fields by gathering all fields found in the sources
outflds = list()
for flds in source_flds_lists:
for f in flds:
if f not in outflds:
# add any new fields as we find them
outflds.append(f)
else:
# predetermined output fields
outflds = header
yield tuple(outflds)
def _standardisedata(it, flds, outflds):
# now construct and yield the data rows
for row in it:
try:
# should be quickest to do this way
yield tuple(row[flds.index(f)] if f in flds else missing for f in outflds)
except IndexError:
# handle short rows
outrow = [missing] * len(outflds)
for i, f in enumerate(flds):
try:
outrow[outflds.index(f)] = row[i]
except IndexError:
pass # be relaxed about short rows
yield tuple(outrow)
# wrap all iterators to standardise fields
sits = [_standardisedata(it, flds, outflds) for flds, it in zip(source_flds_lists, its)]
# now determine key function
getkey = None
if key is not None:
# convert field selection into field indices
indices = asindices(outflds, key)
# now use field indices to construct a _getkey function
# N.B., this will probably raise an exception on short rows
getkey = operator.itemgetter(*indices)
# OK, do the merge sort
for row in shortlistmergesorted(getkey, reverse, *sits):
yield row
