We require the full path to the file where the metadata
will be written to. Let's not make guess...

There were some off-by-one errors and casting bugs.
I should add more tests covering this.

Remove an extra variable we don't need.

Clean up some code by adding the 'Async' suffix to
async methods, removing dead code, and ensuri...

As per code comment - if ReceiveYAsync completes synchronously
it will immediately call SendAsyn...

This has an API break (methods were removed).

Just use a simple IList<IList<string>>, or IList<string>
for these. No need to have funky wrappers!

If we call StopAsync while actively downloading we may
be writing to the file when we go to try ...

We should update the offset *after* we have finished
hashing data. This ensures that anything ca...

This approach is threadsafe and simple.

The main goal of this is to make ITorrentFile and Torrent immutable.
It also clearly separates t...

Run CancellationToken through the tracker APIs. Add a cleaner
API for dynamically adding/removin...

If dozens, or hundreds, of torrents are started
concurrently then we should limit the announces
...

We can now cache a buffer to re-use for our receive loop.
This reduces the amount of times we ha...

This is a small tweak to allow significantly less context
switching while sending messages to a ...

It's a little more efficient this way as we're reading from the socket
in parallel with actually...

Buffers used to send/receive from sockets should be
stored separately to ones used to retain blo...

This is a bit better performance wise as the final
hash will be performed while on the threadpool.

Ensure we have *either* an 'announce' key or an
'announce-list' key.

If all the tiers are empty...

This unblocks the disk io loop so it can process more
reads/writes concurrently.

Spin up a few monotorrent instances and have them connect to
each other. It's the easiest way to...

We still need to pass the 'Peers' list into the event args.

Oops.

As this is controlled via an IDispoable now, we don't need
to verify the correct buffer is being...

Now we're consistently on the threadpool regardless of
whether we need to test an encrypted conn...

Use the 'Async' ones nowadays.

Now all the message encoding/decoding, encrypting/decrypting
as well as the actual network io ha...

We can drop them a little faster if we check peerid
while deciding if we should re-use the conne...

Make it internal, and now that it's internal API we can simplify
it. We only ever want SHA1 and ...

The async methods are fairly awful perf-wise and appear
to do a lot of work synchronously (accor...

If we are listening at an endpoint which ends with
'/announce/' then we should additionally list...

This breaks the API for anyone implementing a custom
connection. They'll need to return a Reusab...

This is a relatively small change in the grand scheme
of things, but it allows for a few improve...

Make BufferPool return an object which allows us to
cache the SocketAsyncEventArgs alongside the...

If you have not called WaitForMetadataAsync and try to create a
stream, you must have passed in ...

Fixes https://github.com/alanmcgovern/monotorrent/issues/314

Patch suggested by usikpavel

Now we can wrap buffer usage in a 'using' statement in the
majority of cases, which is a neater ...

We cannot have proper piece picking in metadata mode as we do
not acutally know how many pieces ...

This should be created inside PieceManager like everything else.
This way when we swap between m...

As soon as we create a PeerId we put it in the Connected list.
This means that if we are in meta...

This reduces the chance of a collision when tests run
in parallel.

We should ignore them when they're received. This is
implemented by special casing the situation...

If you create a torrent from a magnetlink with trackers,
ensure those trackers are maintained as...

Now they can be trivially executed under .NET Core as
well as desktop .NET 4.7.2.

Also, the HttpTracker tests are flakey under .NET Core
as it appears to have a buggy implementat...

In the event there's a bug we may cancel the tcs in the test. When
this happens we don't want to...

Sometimes we *want* to time out and do something with the
exception, so don't force tests to fai...

We can handle a null IDisposable :)

This typically fastpaths and exits early, so make it a ReusableTask
so it becomes zero-allocatin...

Save some repeated allocations by caching the array used
to hold the infohashes. We are likely t...

We should throw the pending exception after the args
have been prepared for reuse to avoid an in...

It's needed in a few places, so ensure we set it consistently
and can easily tweak it in future.

If we receive a libtorrent extension handshake with no value
set for the max pending requests, i...

Ensure the incremental hasher takes the torrent manager
as well as the piece index into account....

The state changed event is asynchronous.

Fixes https://github.com/alanmcgovern/monotorrent/issues/287

This support is built on top of Mono.Nat but can be swapped
to any other implementation if neede...

Greatly simplify the code in ChokeUnchokeManager. We now run a review every minimumTimeBetweenRe...

This is currently slower than regular seeding mode.
I'd like to fix this to avoid frequent choki...

If the TorrentSettings has set the upload slots to 0 it
is supposed to mean 'unlimited'.

This c...

Remove lots of dead commented out code.

This is better than Task.Run and then calling a
blocking API to download the data.

Simplify the code by making it throw an exception
if someone uses a 2 digit version number. As w...