This edits the third party SCTP library, but upstream has already applied this change to master
MozReview-Commit-ID: ERpMc8EvYZ7
--HG--
extra : rebase_source : cea0d3758275b73a395ad2738edd8eb57c833e1a
This allows sending and receiving arbitrarily (we limit to 1 GiB atm) sized
messages while not relying on the deprecated PPID fragmentation/reassembly
mode. The code already supports the ndata extension but it's not activated,
yet. Without the SCTP ndata extension, a large data channel message will
monopolise the SCTP association. While this is a problem, it is a temporary
solution until the extension is being activated. Keep in mind that every
application that uses data channels currently does fragmentation/reassembly on
application-level and it's unlikely that this will change until the popular
implementations (libwebrtc) implement EOR as well. Moreover, until the WebRTC
API specifies an API that hands over partial messages, doing application-level
fragmentation/reassembly is still useful for very large messages (sadly).
We fall back to PPID-based fragmentation/reassembly mode IFF a=max-message-size
is not set in the SDP and the negotiated amount of SCTP inbound streams is
exactly 256. Other implementations should avoid using this combination (to be
precise, other implementations should send a=max-message-size).
It also changes behaviour of RTCDataChannel.send which now raises TypeError in
case the message is too large for the other peer to receive. This is a
necessity to ensure that implementations that do not look at the EOR flag when
receiving are always able to receive our messages. Even if these
implementations do not set a=max-message-size, we use a safe default value (64
KiB, dictated by the spec) that every implementation should be able to receive,
with or without EOR support.
* Due to the use of explicit EOR, this required some major refactoring of all
send-related and deferred sending functions (which is now a lot less
complex). There's now only one place where `usrsctp_sendv` is being used.
* All data channel messages and DCEP messages will be sent without copying them
first. Only in case this fails (e.g. usrsctp's buffer is full), the message
will be copied and added to a buffer queue.
* Queued data channel messages will now be re-sent fairly (round-robin).
* Maximum message size and the PPID-based fragmentation are configurable using
about:config (media.peerconnection.sctp.force_ppid_fragmentation and
media.peerconnection.sctp.force_maximum_message_size).
* Enable interleaving of incoming messages for different streams (preparation
for SCTP ndata, has no effect until it is enabled).
* Enable interleaving of outgoing messages (disabled if SCTP ndata has not been
negotiated).
* Add pending messages flag to reduce performance impact from frequent calls to
SendDeferredMessages.
* Handle partial delivery events (for cases where a partially delivered message
is being aborted).
* Close a data channel/the connection in case the message is too large to be
handled (this is only applied in cases where the remote peer ignores our
announced local maximum message size).
* Various size_t to uint32_t conversions (message length) and back should be
safe now.
* Remove aUsingDtls/mUsingDtls from DataChannelConnection.
* Set maximum message size in SDP and in the data channel stack.
* Replace implicit NS_ENSURE_*'s with explicit NS_WARN_IF's.
* Add SetMaxMessageSize method for late-applying those signalling parameters
when a data channel has been created before the remote SDP was available.
* Limit remote maximum message size and add a GetMaxMessageSize method for a
future implementation of RTCSctpTransport.maxMessageSize.
MozReview-Commit-ID: FlmZrpC5zVI
--HG--
extra : rebase_source : 54e1b838c788a3abbded4fb32fe7c2788f8a9bc0
We have a minimum requirement of VS 2015 for Windows builds, which supports
the z length modifier for format specifiers. So we don't need SizePrintfMacros.h
any more, and can just use %zu and friends directly everywhere.
MozReview-Commit-ID: 6s78RvPFMzv
--HG--
extra : rebase_source : 009ea39eb4dac1c927aa03e4f97d8ab673de8a0e
The new name makes the sense of the condition much clearer. E.g. compare:
NS_WARN_IF_FALSE(!rv.Failed());
with:
NS_WARNING_ASSERTION(!rv.Failed());
The new name also makes it clearer that it only has effect in debug builds,
because that's standard for assertions.
--HG--
extra : rebase_source : 886e57a9e433e0cb6ed635cc075b34b7ebf81853
The patch is generated from following command:
rgrep -l unused.h|xargs sed -i -e s,mozilla/unused.h,mozilla/Unused.h,
MozReview-Commit-ID: AtLcWApZfES
--HG--
rename : mfbt/unused.h => mfbt/Unused.h
This patch makes the following changes on many in-class methods.
- NS_IMETHODIMP F() override; --> NS_IMETHOD F() override;
- NS_IMETHODIMP F() override {...} --> NS_IMETHOD F() override {...}
- NS_IMETHODIMP F() final; --> NS_IMETHOD F() final;
- NS_IMETHODIMP F() final {...} --> NS_IMETHOD F() final {...}
Using NS_IMETHOD is the preferred way of marking in-class virtual methods.
Although these transformations add an explicit |virtual|, they are safe --
there's an implicit |virtual| anyway because |override| and |final| only work
with virtual methods.
--HG--
extra : rebase_source : 386ee4e4ea2ecd8d5001efabc3ac87b4d6c0659f
This patch makes most Run() declarations in subclasses of nsIRunnable have the
same form: |NS_IMETHOD Run() override|.
As a result of these changes, I had to add |override| to a couple of other
functions to satisfy clang's -Winconsistent-missing-override warning.
--HG--
extra : rebase_source : 815d0018b0b13329bb5698c410f500dddcc3ee12
This removes the unnecessary setting of c-basic-offset from all
python-mode files.
This was automatically generated using
perl -pi -e 's/; *c-basic-offset: *[0-9]+//'
... on the affected files.
The bulk of these files are moz.build files but there a few others as
well.
MozReview-Commit-ID: 2pPf3DEiZqx
--HG--
extra : rebase_source : 0a7dcac80b924174a2c429b093791148ea6ac204
Until da6f8ba in upstream git repository, sctp_init_ifns_for_vrf() was
using the MALLOC and FREE macros, using respectively HeapAlloc and
HeapFree. da6f8ba changed the allocations to use GlobalAlloc, but didn't
change the deallocations to use the symmetric GlobalFree.
This doesn't seem to cause direct problems, but when wrapping one family
of allocation functions, the dissymmetry causes allocator mismatch
problems.
Moreover, the MALLOC macro being unused, it might as well be removed,
along the FREE macro, so that both allocations and deallocations use one
API explicitly.
See https://github.com/sctplab/usrsctp/pull/92
Andrea Marchesini
Randell Jesup
Chris Manchester
Tom Ritter
Lennart Grahl
Kartikaya Gupta
Sylvestre Ledru
Kershaw Chang
Bill McCloskey
Carsten "Tomcat" Book
Bevis Tseng
Wes Kocher
Nils Ohlmeier [:drno]
Tom Tromey
Markus Stange
Andi-Bogdan Postelnicu
Tom Schuster
Michael Layzell
Nicholas Nethercote
Kan-Ru Chen
Jinank Jain
Byron Campen [:bwc]
Jan Beich
Mike Hommey