To do list

[donmerendolo]

I'll probably do these this summer:

• Separate the code into several, more manageable and readable files.
• Rewrite main so it's more readable.
• Rewrite _fetch_all_items so it doesn't take two kinds of data.
• Add users option to config.cfg.
• The program still creates legacy uris and external_ids in config by default. It shouldn't do that.
• Running a bare -l changes the config file unnecessarily.
• In config.cfg, format can take json and csv separated by a space. fields takes multiple arguments separated by commas. Choose one or allow both.
• Write some tests.
• Publish in PyPI? Use $XDG_CONFIG_HOME and place .cache somewhere else?
• fields option in config should, by default, have the default fields, not be empty. Probably a comment above it with all options too.
• Binaries' names are currently kinda ugly.

I'll probably release all of these in a v1.0.0 version, making it as much backwards compatible as I can.

Fable had some interesting suggestions and bugs to fix too:

Details

Had a look through `exportify-cli.py` — a few things you might want to add:

Actual bugs I spotted:

• The output option in config.cfg is dead code: -o has default="./playlists" in the click decorator, so output_param is always truthy and output_param if output_param else cfg.get(...) never falls back to config. Same pattern as your -l issue — config/CLI resolution deserves one unified pass.
• Sorting is lexicographic on str(): --sort-key Popularity gives 9 > 10 > 100 order, and str(None) or "" is "None", so the or "" fallback never fires. Numeric keys need numeric comparison, and Nones should sort predictably.
• os.get_terminal_size() raises OSError when stdout isn't a tty, so exportify-cli -l | grep foo or running from cron crashes. shutil.get_terminal_size() has a fallback built in.
• In _fetch_all_items paginated mode, total = results.get("total") can be None for some endpoints, and while len(items) < total would then throw a TypeError. Worth a guard (results.get("next") is the more robust loop condition anyway).
• The format resolution uses substring checks ("csv" in cfg.get(...)), so a typo'd or empty format silently exports nothing with exit code 0. Validate and error instead.
• If you're building with PyInstaller --onefile, Path(__file__).parent points to the temp extraction dir (_MEIPASS), not next to the binary — so .cache and config.cfg may be written somewhere ephemeral. Check for sys.frozen and use Path(sys.executable).parent in that case (this ties into your XDG item).

Smaller stuff / polish:

• CSV output writes list fields as Python reprs (['Artist A', 'Artist B']). Original exportify joins with commas — probably want ", ".join(...) for CSV at least.
• README says playlist names support "partial matching" but the code only does startswith prefix matching. Align one with the other.
• -c pointing at a nonexistent explicit path silently starts the config-creation wizard there; arguably that should be an error when the user passed the path themselves.
• get_version() shells out to git at import time on every invocation — cheap but unnecessary; lazy-evaluate it.
• The redundant if unfetched_ids: check inside the show-fetching loop in _fetch_all_items (it's already iterating over it).

Feature ideas:

• Skip-unchanged exports: compare the playlist snapshot_id against the last export and skip if identical — makes -a on a cron job fast and diff-friendly (the original exportify does something similar).
• Non-zero exit code / summary when some requested playlists failed or were skipped (ambiguous prefix currently just prints "Skipping" and exits 0).
• Optional playlist metadata in output or a sidecar file: description, owner, snapshot_id, export timestamp.
• A --quiet/--verbose flag now that you have logging set up (currently logger.info is invisible).
• Once you split into modules and add tests: ruff + a CI lint/test workflow, since you already have .github/workflows.

The config/CLI resolution one is probably worth prioritizing since three of your existing items (-l writing config, legacy uris/external_ids, format vs fields syntax) plus the dead output fallback all stem from the same tangle — a single "config schema + precedence" refactor would knock them all out together.