I've committed a fix upstream that causes undeliverable-here messages to be rejected: https://svn.apache.org/r1715199. I believe this avoids any violation of the spec.

If delivery-failed is false (and the message is not rejected) then the delivery count does not appear to get incremented as things stand.

Reproducer:

#!/usr/bin/python
import optparse
from proton import Delivery, symbol
from proton.handlers import MessagingHandler
from proton.reactor import Container



def dump(obj):
  for attr in dir(obj):
    print "obj.%s = %s" % (attr, getattr(obj, attr))


class Recv(MessagingHandler):
    def __init__(self, url, failed):
        super(Recv, self).__init__(auto_accept=False)
        self.url = url
        self.failed = failed

    def on_start(self, event):
        event.container.create_receiver(self.url)

    def on_message(self, event):
        print event._attrs
        event.delivery.local.failed = self.failed
        event.delivery.local.undeliverable = True
        event.delivery.update(Delivery.MODIFIED)
        event.delivery.settle()
#        event.connection.close()

parser = optparse.OptionParser(usage="usage: %prog [options]")
parser.add_option("-a", "--address", default="localhost:5672/examples",
                  help="address from which messages are received (default %default)")
parser.add_option("-f", "--failed", action="store_true", help="delivery-failed")
opts, args = parser.parse_args()

try:
    Container(Recv(opts.address, opts.failed != None)).run()
except KeyboardInterrupt: pass


1) service qpidd restart; qpid-send -a "examples; {create:always}" -m1
2) python modified.py
3) service qpidd restart; qpid-send -a "examples; {create:always}" -m1
4) python modified.py -f

Expected behaviour:
2) prints one message like:
{'message': Message{priority=0, first_acquirer=1, user_id="", content_type="text/plain", properties={"sn"=1, "ts"=1448228051627021916}}}

4) prints the same

Observed behaviour:
2) prints same message like:
{'message': Message{priority=0, first_acquirer=1, user_id="", content_type="text/plain", properties={"sn"=1, "ts"=1448228301965840558}}}
..

forever

4) prints messages like:
{'message': Message{priority=0, first_acquirer=1, user_id="", content_type="text/plain", properties={"sn"=1, "ts"=1448228254722139242}}}
{'message': Message{priority=0, delivery_count=1, user_id="", content_type="text/plain", properties={"sn"=1, "ts"=1448228254722139242}}}
{'message': Message{priority=0, delivery_count=2, user_id="", content_type="text/plain", properties={"sn"=1, "ts"=1448228254722139242}}}
..

forever

(In reply to Gordon Sim from comment #1)
> I've committed a fix upstream that causes undeliverable-here messages to be
> rejected: https://svn.apache.org/r1715199. I believe this avoids any
> violation of the spec.

Although it is not stated there that undeliverable-here=true MODIFIED messages should be offered to other consumers, I expect that is the intention behind the spec. But I agree implementing it would be a bigger task - every message would have to keep a list of consumers setting undeliverable-here in past..

I.e. I am not happy from the implementation but dont see much better.

> 
> If delivery-failed is false (and the message is not rejected) then the
> delivery count does not appear to get incremented as things stand.

Agreed from my experiments, clarifying with original reporter their reproducer/scenario.

This bug was fixed upstream (QPID-6870, as much as we will fix it) as of 1.35.0.  Thus it was part of the MRG release downstream when we rebased to qpid-cpp 1.35.0.  So closing since it's now in MRG.