Factor duplicate serial.Serial initialization out of Windows and POSIX serial port modules

[Jean-Paul Calderone]

twisted/internet/_posixserialport.py and twisted/internet/_win32serialport.py both have a BaseSerialPort subclass. Each subclass creates a serial.Serial instance in basically the same way. This isn't a lot of duplicate code, but it's a little. The main argument in favor of removing the duplication is that if a single method were responsible for this, there would be an obvious place to put documentation for the parameters.

[Jean-Paul Calderone]

(In [29429]) Branching to 'refactor-duplicate-serialport-4525'

[Jean-Paul Calderone]

I mostly just pulled some code out of the #3802 branch. Note in particular how timeout is handled in the affected code.

Also, I don't know how well the new API docs are going to be rendered. The serialport modules are constructed in a weird way, probably confusing to pydoctor.

​build results

[Lucas Taylor]

Minor modifications

[Lucas Taylor]

Attaching some minor modifications:

• Added reactor and protocol to the BaseSerialPort initialization and set attributes there since they are common to both implementations. Also, the way BaseSerialPort was being initialized would have resulted in an error due to parameter order.
• Initially flush serial port in BaseSerialPort
• Removed an extraneous protocol attribute assignment in _win32serialport.py
• docstrings for requirements pyserial and pywin32

[Jean-Paul Calderone]

Attaching some minor modifications

Thanks. I'll check this in. Meanwhile, I think #3690 should be reviewed and resolved first, since it introduces the first serial port tests we'll have, and refactoring without unit tests is crazy. There'll be some minor conflicts, easily resolved.

[Thijs Triemstra]

Those pyserial requirements should probably go in the readme file, not in the module.

[Jonathan Lange]

Thanks for removing the duplication.

I don't know whether I agree with thijs's comment, that is, I don't know what the correct principle is. From a pragmatic POV, I think the dependency requirements are fine where they are.

This bit:

+    Non-implementation of the serial port transport, only actually used when
+    running on a platform for which Twisted lacks serial port support
+    (currently POSIX and Windows on CPython are supported).

seems a bit clunky, but I can't think of any improvements.

In a nutshell, I'm fine for this to land as-is, pending #3690.