- Improve the AudReader and WavReader by performing fewer, larger stream reads to consume bytes more efficiently.
- Improve ImaAdpcmReader.LoadImaAdpcmSound by accepting an output buffer rather than allocating a new array.
- Improve StreamExts.ReadAllBytes by using MemoryStream.CopyTo. This internally rents a much larger buffer from the array pool, allowing for fewer, larger stream reads.
Previously, the MiniYaml.From* helpers such as FromStream would consume the entire input and then return a list of top-level nodes. Now, the input is processed using deferred execution and top-level nodes are yielded as they are resolved from the input.
The motivating use-case is MapCache, which currently manually buffers nodes before passing to MiniYaml.FromString in order to improve responsiveness when large payloads are processed. Now that MiniYaml.FromStream yields results back as they come in, we can switch to that without disadvantage.
The maintains the performance where map cache can update search results as each node comes in over the network rather than having to wait for the entire batch to be transferred. Now we can also remove the string buffer that captures each node, reducing memory pressure and simplifying the code.
Also avoid ReadBytes calls that allocate a buffer by either updating the stream position (if not interested in the bytes), by reusing an input buffer (if interested in the bytes), or using a stackalloc buffer to avoid the allocation (for small reads).
- Stream lines in as memory rather than needing to realise a string for each line, via a new method in StreamExts.
- Use span to avoid string allocations during parsing until we want to realise the node itself, in MiniYaml.FromLines.
- Change several callsites to use the streaming extension method rather than string method where possible.
Change callers of this method to use the ReadBytes extension method that ensures the requested amount of bytes are consumed.
Also tighten argument validation.
- Almost all calls to Stream.Read were broken. These have been patched to all go through ReadBytes which itself has been fixed to function correctly. The key thing to note is that Stream.Read is very much allowed to return less than the requested number of bytes. If this happens and you're not checking the return result, you'll be working with partially initialized arrays and really bad stuff happens when you do that.
- Call CopyTo rather than copying between streams manually.
- Peek and ReadUInt8 have been changed to avoid a pointless array allocation which is significant overhead for such simple calls.
