We can probably rely on an interceptor (PiecePickerFilterChecker)
and a StandardPicker to fulfil...

StreamingPicker can now be implemented as a PiecePickerFilter
rather than a full PiecePicker. It...

The idea is to:
1) Make it easier for the engine to indicate when a peer should be
used to downl...

The return value is guaranteed to be immutable (again).
The previous commit returned the input p...

When we seek to a new location in the piece picker we want
to cancel any pending requests and im...

There's (currently) only one way to request or cancel
a piece, and that's calling CreateRequest ...

It's helpful if you ever want to compare against the
list of ConnectedPeers.

If a peer sends data slower than 1 block every 15
seconds we should close the connection.

If we download just the last piece it may not actually be enough
data. The last piece could be a...

This will make things a little faster maybe.

Improve 'ContinueAnyRequest' to support prioritising the
high priority pieces over low priority ...

The streaming picker is intended to be forward-only, so we
should exclude pieces which come befo...

If a peer does not send the 'MetadataSize' key it
means they do not have metadata to send us. We...

It's not impossible for the connection to be disposed
before we reach the Socket.Connect call.

...

The encryption spec does not allow the initiator of the
connection to set a preference. As such,...

If a block is abandoned, only peers which have not been involved
in hash fails should be allowed...

It's simpler all round to mark pieces as abandoned instead of
monitoring this on a block-by-bloc...

If we request a piece and it is sent at less than 1kB/sec we
should simply disconnect the peer. ...

This is less efficient as data can arrive out-of-order which
means we'll write data to disk, the...

We know the peer will not supply any more data if they
choke us, so we can treat those pieces as...

They were annoying me :p

Don't write all log statements to the console in release
mode as it makes it a bit harder to per...

The DiskManager implementation may, or may not, use a
memory cache. As per comment in the code -...

It would be incredibly complex to support a scenario where some
parts of a single read are in th...

We shouldn't rely on the media player receiving the data
to seek to the start of the stream befo...

Some APIs deliberately read until 0 is returned, such
as Stream.CopyToAsync.

If the peer who was downloading a piece has been disconnected,
treat the piece as abandoned and ...

If we seek to the same piece we're already in then we don't
need to cancel pending piece requests.

The tests don't necessarily run a main loop which ticks
the DiskManager to process queued writes...

When we've downloaded all the data we intend to, flush the
data to disk before changing to 'Seed...

We cannot iterate over 'CachedBlocks' and use
an await unless we write our code so that it
is aw...

Don't bother creating a real Udp based DhtListener for the tests
as we never acutally use it.

T...

Use two queues to allow tasks to be processed with
fewer locks.

By swapping the queue object we...

Change around ownership of the Semaphore used to control
access to the streams (again).

There w...

When we use a Stream we now update it
in the usage order :)

If we have a piece in the disk cache and we 'read' it,
we can be reasonably certain the engine w...

Defaulting to using a MemoryCache should improve performance
when Blocks arrive slightly out of ...

The new semantics make it better to enter endgame mode when
ContinueAnyExisting returns null rat...

Now we only try to continue an existing request if the
peer has been unable to select one normal...

This allocates far less than the built-in implementation
so it's a net win for monotorrent. The ...

This makes it less likely there'll be a bug later on.

Urgh. Oops.

Thanks to jpmikkers

The filestreams we use to read/write data from disk are now
protected with just a SemaphoreSlim....

This is to ensure that alpha versions are strictly monotonic
when sorted as strings, as per spec.

Also a bunch of misc cleanups to make the whole thing work.
I was too lazy to proper separate ou...

If the Settings objects were immutable and users had to
pass in a new Settings object whenever t...

We should ensure we're executing on the IO thread
when we're interacting with the filestreams.

...

We can start issuing unstable nuget packages for
people who want to live on the edge!

Will have...

The API expects this to be the path to a specific file,
so when we autogenerate a value ensure w...

Co-authored-by: guangzhi.wu <[email protected]>

If a BEncodedDictionary has been encoded with the keys in the
wrong order, we will now compute t...

We should tie the lifespan of the local peer listener
to the engine it's registered with.

If no...

Add an early exit to this task. If we find at least 128 peers
we should stop sending further que...

It's better to add peers to the table by virtue of them responding
to a message we've sent. Do n...

We now allow the 'Initialization' task to indicate it has
completed before it has fully inserted...

Have one codepath for initialization which ensures the
state is updated consistently.

When we send a FindNode message to a torrent, we do
not want to add *every* node to the DhtTable...

This clearly needs better tests, or even any tests at all :p

Improving the integration tests to...

Every await can add overhead, so let's avoid
awaiting something if we don't need to!

It's a little simpler than the original code which
used a manualresetevent.

The tests cover this completely so we can simplify
things by removing the code which attempts to...

Nothing uses this nowadays

Firstly, don't reset the stream every time we connect
to a peer who supports metadata mode. That...

Previously we'd emit the log line
'connected to self' if we actually had too many
active connect...

I only really need net472 and net2.1. That'll test all
relevant codepaths. Might as well include...

Ensure the packagereferences are restricted
to use a version of the nuget dependencies
which are...

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