Use new, flat API for SSL peer name

Added tag rabbitmq_v2_8_2 for changeset 027349f33211

Junk bug24866

API change

Earlier versions of Erlang don't have integer_to_list/2 auto-imported.

Merged bug24763 into default

Destination on MESSAGE frames reflects SEND destination, not SUBSCRIBE destination;
Treat default exchange and amq.topic exchange specially;
Escape content;
Adjust tests to cope with new expectations.

Merge bug 23463

Generalise memory alarm

Merged bug24799 into default

Merge in default.

Update connect_options.py to check for error frame (if bad connect)
and to set default user before test (so it passes).

Added tag rabbitmq_v2_8_1 for changeset 9c0ef158faf8

Added tag rabbitmq_v2_8_0 for changeset f98894d59e3c

Merge bug24796

Test implicit connections even if we *are* doing SSL tests, by turning implicit connect on just for the test.

Removed impossible not_found in login_header/3

Log the various authentication failures possible.

Updated test ssl.config to remove default user (and implicit connect);
adjusted tests to avoid using implicit connect everywhere;
put check in login_header to return undefined if lookup returns undefined.

Merge bug24791 into default

Slightly more comment

Add failing test. We can't implement this in Python since stomp.py seems buggy in this area anyway (I'm not sure it sends DISCONNECT).

Make the reader transient. Explain why.

Make sure STOMP connections have a name, and thus fix mgmt connection listing.

Merge bug24761 into default

Allow "server:.*" header on CONNECTED frames in tests.

"server" header in CONNECTED frame.

Consistency.

Merge bug24753 into default

Merge bug24182 into default

Merge in default (something of a biggie, merge of bug24744 demanded we move stuff around).

Merge bug24744

Cosmetic

Merge stomp_processor_sock to stomp_reader

implicit return type for start_processor

rabbit_net:send was wrapped in catch previously

SendFunc shall be SendFun

Allow the protocol name to be customized

Merge bug24542 into default

Some renaming, macros and refactoring in the name of encapsulation.

Merged bug24751 into junk

Rename internal subscription prefix

Oops

Rename internal subscription prefix, add safety checks

Forgot to add this...

Revert accidental inclusion of bug24753.

"Promote" implicit_connect to the top level since it is no longer just for default_user

Move the mode thing into the server too.

SSL client certificate authentication. In theory based on Shane Hathaway's patch but there's probably nothing left of it by now.

Rebuild the test config every time and thus make it a lot damn clearer whether or not the SSL tests are running and why.

Cosmetic: the power of pattern matching

Get heartbeats working again - must support both sync and async data sending

Don't enter an infinite loop if implict connect fails.

Subtle typo

Move Sock functions out of processor

Unused variable warning

Remove code that checks for config vars not being set: this can never happen.

Despaghettifying part 1 - move most of the rabbit_net calls to init

Patch received from Shane, updated to be branched from default.

Backed out changeset c0b6b23c6ec9

Use unpredictable subscription id

Provide fake subscriptions for reply messages

missed io:format

Modularize rabbit_net calls in stomp_processor - allow pluggable transports other than raw tcp.

Merge 24740 into default.

Redundant.

ensure that generated queue names meet the AMQP spec

there is no need to hash the subscription id
doing so doesn't make it any more unique or secure

produce valid utf8 in generated queue names

Merged bug24505 into default

Merged bug24670 into default

Merge bug24606

This is obsolete, delete rather than update copyright.

Update copyright 2012

refactor: use new rabbit_misc:format/2

Merge in default

Update copyright 2012

rabbit_guid:string_guid/1 -> rabbit_guid:string/2.

merge bug24336 into default

oops

merge default into bug24336

extract error logging
so we don't conflate it with returning an {error, ...} tuple

Don't explode on AMQP death.

Merge bug22470

Fix the little test client so it actually uses the parser correctly and does not drop messages.

Actually pay attention to the rabbit_stomp_frame:parse/2 API.

Merge default

Whitespace

produce only one log entry for abnormal connection closure
instead of two

merge default into bug22470

Tweak the tuple we get...

Only link to the connection, otherwise we get a race where the link conn -> stomp_processor -> channel can take out the channel from under the connection. This solves bug24336 but is perhaps not completely ideal.

Merge bug24698 into default

Merge default in

Merge bug24688 into default.

Merge in default

use the existing error reason formatter

refactored error message creation
and got rid of the trailing ~n - the message does *not* get formatted
by Erlang, and the 'message' is meant to be a *header*, so, ideally,
on one line.

merge default into bug24336

more accurate log message/level

always use rabbit_log for logging

use the new rabbit_log:log for STOMP connection logging

error_logger flood prevention

consistency

use the new rabbit_net:connection_string/2
and bring log messages in line with rabbit_reader

merge default into bug24698

make start_client/2 return the reader instead of the sup
...since it is meant to be returning the process controlling the
socket.

In practice using the sup was fine since it would usually terminate at
the same time as the reader.

Also, some cosmetic changes to align the code with
rabbit_networking:start_client/2.

move ssl upgrade from acceptor to reader
...thus allowing further connections to be accepted while the upgrade
takes place.

The reader & processor startup needed to be re-arranged. The processor
needs to know about the socket, but that now doesn't become available
until the reader has upgraded it. Hence we can no longer start the
processor at the same time as the reader. Instead we get the reader to
start the processor.

We also need to change the reader's restart type from 'temporary' to
'intrinsic'. The former was always a bit dodgy since it meant a reader
exit would not terminate the supervisor, thus leaking a process. In
practice that was generally prevented due to the reader containing a
'try ... after' that would terminate the processor, which is an
'intrinsic' process and hence would take down the supervisor. But the
reader can exit before entering the 'try ... after', e.g. if
rabbit_net:peername fails (rare, but certainly possible in cases where
a client connects and immediately disconnects), and, with the changes
in this bug, when the ssl upgrade fails.

use new rabbit_networking:tcp_listener_spec/6

keep up with broker API changes

refactor: eliminate pointless indirection

Sicne we trap_exits why don't we link to the connection / channel(s) and do something sensible if they die?

Silence warnings

The other plugins do not announce themselves like this, stomp should not either.

only create destination queues once per session
...which is more efficient that doing so once per SEND.

Continue on our quest towards being able to count.

Mask lumpiness

Turns out I actually can't count. Previous version assumed one send stats for every recv stats, but the recv ones tend to run behind...

WIP

A simple publish-and-consume-as-fast-as-you-can test since I'm not convinced about the performance of some of the clients we've poked at. I was going to call this multicast_plain but I think puns like that would be discouraged.

Added tag rabbitmq_v2_7_1 for changeset 3c6e64157010

Add test to detect this problem

Merged bug24364 into default

Wrap

Put description in test.

Merge bug 24572.

Simplify take_prefix to be more idiomatic Erlang

Simple test for rejection.

Added check for headers not allowed on SEND
o also factored out some more headers

Factored out header strings.

Cosmetic

Removed take_prefix/3; simplified take_prefix/2.

Rewrite split_* to split/2; taking Splitter list as parm.
o introduce take_prefix/2 to check for prefix AND remove it combined
o take_prefix/2 also used to replace strip_leading_blank logic

Tidied and corrected split functions:
o renamed slash_split/1,3 to split_content/1,3;
o removed strip_leading_blank (effect now in strip_content/1 matching);
o removed push_non_null_rstr/2;
o made both splits preserve order;
o splits now treat null string correctly.

Recoded regexp:split and regexp:gsub calls to avoid regexp.

Merged bug24541 into default

Whitespace

Moved common constraints into an hrl

Code to pass test.

Add test to demonstrate problem.

Added tag rabbitmq_v2_7_0 for changeset f52d6ad39a3e

Merge bug21319 (include plugins in source and binary packages)

update installation instructions

stomp needs amqp_client

Merge bug24426 into default

Silently cope with unexpected conserve memory calls.

Merging heads

Added tag rabbitmq_v2_6_1 for changeset d21e2feffbf6

cosmetic and minor refactors

Added tag rabbitmq_v2_6_0 for changeset 15d580297dbe

merge heads

Added tag rabbitmq_v2.6.0 for changeset 15d580297dbe

Merge bug24375 into default.

Update close_socket in rabbit patch for test.

Merge in default.

Merge bug24373 into default.

Merge in default

Handle case when we get acks with no recorded receipts

Merge bug24312 into junk

Fix handling of confirms for unrecorded receipts

Merge with default

Merge bug24324 into default

Support header trimming for 1.0 clients

Rename and simplify shutdown_channel_and_connection to close_connection.

Remove catch from rabbit_net:*name calls but retain case (in case of {error, einval}).

Merge with default

Merge bug24324 into bug24312 to check fix.

Backout rabbit.patch alterations

Merge bug24366

Merge default

Make shutdown_channel_and_connection more comprehensive

Merge default

Prevent premature death of stomp_reader and stomp_processor to clean up connection:
- change reader to temporary from intrinsic
- protect alien calls from processor by catch so as to avoid premature death
- call processor:flush_and_die/1 from reader when mainloop exits before reader dies

Merge bug24186

Remove default prefetch for /queue destinations

Merge in default.

Improve ack tests

merge default into bug24186

bug24331 merged into default

Added one more test - no content length + binary (utf) data on the wire.

Merge bug24304 into default

Add test to check we do NOT supply a content-type header when omitted

Remove the default content-type header for SENT messages.

Change logic in Connection.close_socket()

Whoops -- adjust again.

Sleep slightly before closing socket to allow test_send_and_disconnect() to work.

Tweak to test to diagnose CI build failures.

A tweak to test to diagnose CI failures.

Make small change to stomp connect.py close_socket() logic.

Upped wait times in three tests to see if CI builds are better. See bug24312.

Merge bug24255 into default.

Adjust wait time again in test.

Merge default in.

Recoded aessertTrue() in test_interleaved_receipt_no_receipt_tx.

Merge bug24257 into default.

Merge in default, resolving conflicts.

Merge bug23864 into default

Fixed heartbeat byte

Support for durable subscriptions

Merge with bug23864