coverage: 96.875%. remained the same
when pulling e6c067e on uiolee:searchparam
into 1ad96ca on hexojs:master.

The code currently uses four sets of URL handling components:

• the functions url.parse() and url.format(), which are deprecated;
• the function querystring.unescape(), which, behind the scenes, calls function decodeURIComponent() and is, according to its documentation, "generally not expected to be used directly" (it is called by url.parse()).
• the functions encodeURI() and decodeURI(), which encode/decode a different set of characters than encodeURIComponent()/decodeURIComponent(), in a different way;
• the classes URL and URLSearchParams.

The functions/classes in each of these groups work fine together but combining them across groups produces unexpected (and sometimes invalid) results because their functionalities do not overlap.

What's the purpose of this decoding and encoding?
It only introduces confusion and bugs.

I would avoid the operations that should cancel each other (encodeURI(decodeURI(...))) and stick to a single set of functions/classes to manipulate the URLs.
Since url.parse() and url.format() are deprecated and encodeURI()/decodeURI() deal with the whole URL at once and do not play well with decodeURIComponent()/querystring.unescape(), I would suggest to use only class URL to validate, parse and possibly re-encode the input URL correctly.
Its documentation describes what it does and it seems to cover all transformations and fixes that function encodeURL() does when it attempts to fix a badly encoded URL.

@axiac

Thanks for your advice.

I also don't know the purpose of some of the code.
They may solve certain problems.
It may also be a historical issue, at least hexo was born before WHATWG URL were added to node.

However, hexo-util is referenced by other packages, so it has to be changed with caution