cli.console: refactor and move into sub-package #6496
Conversation
bastimeyer
force-pushed
the
cli/console/subpkg-and-wrapper
branch
4 times, most recently
from
8194ee6 to
47ffc78
Compare
|
Been struggling with this for a bit, but it's fine now (going to merge Initially, the wrapper class didn't subclass from Then I decided to fix this by first subclassing from TextIOWrapper and passing the buffer and config attributes from the given stream object (a TextIOWrapper) to its super class constructor. No issues were expected and test ran fine locally on Linux and Windows, but the Windows CI runners failed. For some reason, LF chars were translated to CRLF, which didn't make any sense considering that no newline-translation is being done when re-using the wrapped stream's config attributes. After that, I decided to subclass from TextIO instead and define all missing methods+properties from the abstract base class, so mypy is happy. This meant that So now I've reverted back to the first approach but added a respective |
bastimeyer
force-pushed
the
cli/console/subpkg-and-wrapper
branch
2 times, most recently
from
1fb4387 to
901e69d
Compare
- Flush and close console/file output streams manually on process exit - Simplify the `BrokenPipeError` fix, as streams are already flushed
- Add `StreamWrapper` base class which passes through everything to the given `TextIOWrapper` object. Don't subclass from `TextIOWrapper` or `TextIO`. - Wrap and replace `sys.stdout` / `sys.stderr` - Restore on `ConsoleOutput.close()` - Fix typing in `ConsoleOutput`
Move terminal-related functions to `streamlink_cli.console.terminal`
bastimeyer
force-pushed
the
cli/console/subpkg-and-wrapper
branch
from
901e69d to
63ae8b8
Compare
Some more changes in preparation for #6449. Doing this in a sequence of PRs instead of just one giant one...
streamlink_cli.consolemodule into a sub-package and move theConsoleUserInputRequesterinto its own module.ConsoleOutput._write()method(s), so adding themsg_status()method in the future is easier (used for the download progress status line at the bottom)ConsoleOutputStream, which wrapssys.stdout/sys.stderrand simply passes through all attribute lookups. This is necessary so we can intercept writes to stdout/stderr which we don't control via theConsoleOutput, e.g. in Streamlink's dependencies. Otherwise, random writes to these text streams would break the status line at the bottom.streamlink_cli.progressinto thestreamlink_cli.consolesub-packageprogressmodule (again, this has already been moved twice in the past)Uncommitted and unfinished changes on top of this branch that will come in follow-up PRs:
Progressand send messages toConsoleOutput.msg_status()instead of writing to a stream object (sys.stderr) directly and adding a carriage-return character with white-space padding at the end of the lineStreamRunnerand its setup due to theProgresssetup changes (rewriting tests is a bit annoying)ConsoleOutputStream(with subclasses) which handle regular messages and status messages, depending on available terminal features (ANSI control sequences, Windows console, or neither).ConsoleOutputStreamWindowsclass including its detection is still unfinished and currently falls back to the default writer implementation which does not support status messages--progress=forcedoesn't make sense anymore when the required writer features are not supported, as we're now merging the regular and status messages into one stream. An idea I have is making it write the status messages as regular messages, but this would require a slower time interval by theProgressclass in that case, as it updates four times per second. Basically another (minor)Progressrefactor.