Use the default Jieba dict for Chinese search if not set

[Snoopy1866]

Subject: if user do not set the dict option, fallback to the default dict provided by jieba

Feature or Bugfix

• Feature

Purpose

• Provide a fallback if user do not set the dict option when html_search_language value is zh

Detail

• jieba has provided a default dict, whose path can be easily obtained by call jieba.get_dict_file().name

Relates

• https://github.com/fxsjy/jieba/blob/67fa2e36e72f69d9134b8a1037b83fbb070b9775/jieba/__init__.py#L374-L378

[picnixz]

AFAICT, this would leave a file descriptor opened (jieba.get_dict_file() opens a file in read mode). So ideally, we should find a way to retrieve the default dict path without having to open the file. Would it be possible?

[Snoopy1866]

@picnixz yes, because the default dict is installed with jieba, so we can directly get its path using os.path.join(os.path.dirname(jieba.__file__), "dict.txt")

[picnixz]

Can we use the jieba constant of dict.txt instead? (I think there is a constant with that name) or is this constant not meant to be publicly available?

If we want to only rely on public API and not on dunder attributes, we could open the file using a with statement, get the filename and then close it. It would probably be costly on Windows machines because open() calls can be up to 15x times slower compared to Linux but that would probably the second cleanest way to simplify maintenance.

[Snoopy1866]

1. We can use jieba.DEFAULT_DICT_NAME to get the default dict name, but without its parent directory path.
2. We can use inspect.getfile(jieba) to get the path of the module jieba.
3. Finally we can use os.path.join(os.path.dirname(inspect.getfile(jieba)), jieba.DEFAULT_DICT_NAME) to get the path of the path of default dict file, which is not relying on dunder attributes.

One more thing, I find the inspect provide a 'public API' inspect.getabsfile which can get the absolute path of specify module, but it has not documented for 10 years. cpython#56526

If a not documented API is not a public API, it's better to use inspect.getfile instead, I think.

But at the same time, Github Copilot says that inspect.getfile may return a relative path, but I have not found any evidence on the Internet.

[Snoopy1866]

I have modified my code, please review again. @picnixz

[picnixz]

Can we have this import in the function it is being used just to save a bit of import time (unless this is already imported by some other dependency)?

[AA-Turner]

Why use inspect here? It's a very heavy package, we can just use .__file__?

[picnixz]

I wanted to use a public API but maybe it's an overkill (my bad since it was my suggestion :')).

[Snoopy1866]

So just revert to 46753b9 and dynamically get the dict file using jieba.DEFAULT_DICT_NAME is enough, I think.

[Snoopy1866]

Why use inspect here? It's a very heavy package, we can just use .__file__?

Should I modify the code to meet it?

Can we have this import in the function it is being used just to save a bit of import time (unless this is already imported by some other dependency)?

May be that's a compromise, but I already have no better idea about how to make a balance between using public API and avoiding importing a heavy package.

[picnixz]

Let's just use the dunder attribute directly. Sorry for the back and forth

[Snoopy1866]

Let's just use the dunder attribute directly. Sorry for the back and forth

Done! Please review it again.