gh-132631: Fix json.tool when using --json-lines and a file as input

[mpkocher]

• Fixes case where --json-lines and a positional file is provided to json.tool
• Add a specific test for this case

gh-138960: Fix json.tool when using --json-lines 

• Issue: JSON CLI: I/O operation on closed file" when parsing JSON Lines file #132631

[mpkocher]

This is perhaps the most minimal fix, however, there's an alternative version that would be more invasive.

Some observations:

• We should try to use standard python constructs if possible. In this case using context managers would make sure the file handle is closed gracefully.
• We should try to document non-obvious error handling. The ValueError catching isn't particularly obvious (file closing case?). It's also not obvious why main function is doing some error handling/catching, and there's a separate error handling in __main__.
• Not really a fan of the duplicate for obj in objs copy and pasting. Python has functions as first-class citizens, we should be able to write a single for obj in objs: do_stuff(obj) to encapsulate the optional colorization and translating to json functionality.

[picnixz]

This should be a finally I think. If there is an exception happening in json.dumps for instance, we would have it left open. So this should be moved after the except ValueError as e with some finally.

[mpkocher]

I tried looking at the git history of this file and it's a bit tricky to understand why it's not using content managers. At one point it did have context manager approach, then it was migrated to manually calling of open.

If there's a friction point, it's the "in place" feature. I don't think the in place feature/requirement ever fundamentally worked for the json lines case.

[picnixz]

This one can fail directl, so please be careful that we close the infile beforehand (that's the reaason why we have the try-finally in the first place).

[nineteendo]

We're probably using a generator here to save memory. Otherwise we could've just used a list comprehension.

[picnixz]

My comment was about the else branch. It's not a generator, it's a regular tuple.

[hugovk]

See PR #132632 for another attempt.

[mpkocher]

The only test that is failing for me locally is the test_writing_in_place.

The "in place" functionality is a bit surprising and doesn't really have clear requirements around error handling.

It's a bit disconcerting that in the "in place" mode you can overwrite/delete the original file if there's an error.