Changelog

Version 2.1.0

Files are now downloaded into the appropriate subdirectory. For example if you upload parent/child.txt, and then later download that file, the parent subdirectory will be created.
FileSenderClient.download_file now has some additional args: file_size and file_name which will enable better functionality if provided. They are typically obtained from FileSenderClient._files_from_token
All terminal output is now through the logging module. You can use the new --log-level CLI parameter to configure the amount of info that is printed out.
Update the CLI default concurrency to 2 for chunks and 1 for files. This seems to be moderately performant without ever failing

Implement directory tree uploads. You can now pass directory paths anywhere you could previously just pass file paths, including the CLI and the Python API. However, note that currently the directory hierarchy won't be preserved. So if you upload dir_a/file_a.txt and dir_b/file_b.txt, they will simply be downloaded as file_a.txt and file_b.txt with their directories stripped out. This is a limitation of the current API. See https://github.com/filesender/filesender/issues/1555 for context.
A progress bar to uploads

Memory usage in 1.3.0 was incredibly high, because the entire file was being kept in memory due to a bug. This is now resolved

The concurrent_reqs and concurrent_reads client arguments, as well as their corresponding CLI arguments have been replaced by concurrent_chunks and concurrent_files. Read the docs to understand how these work
Automatic retries for transient failures.

Removed the annoying rich error logs which often made it impossible to determine the true underlying error

A bug where the FileSender server would respond with a redirect, which would cause the client to fail

concurrent_reads argument to the FileSenderClient() constructor. This parameter controls the maximum number of file chunks that can be processed at a time.
concurrent_requests argument to the FileSenderClient() constructor. This controls the maximum number of API requests the client can be waiting for at a time.
benchmark notebook that evaluates how manipulating the above two parameters affect runtime and memory usage
A simple test that validates the effectiveness of the above parameters
A filesender.benchmark module which contains utilities for benchmarking that are used by both the notebook and the test

--version flag, to check for the current package version
server-info command, which is a utility for describing the configuration of the server you are using

httpx.PoolTimeout errors, by enforcing a newer httpcore library
json.decoder.JSONDecodeError errors, by not assuming that all errors are JSON. This should better reveal the true underlying error