Should the Document constructor use a different parser for collapse_whitespace? about delb-py HOT 5 OPEN

i've been thinking that whitespace would make a good major topic for the 0.5 version. and, instead of relying on libxml specifics, we can implement serialization natively. that's to be engineered for the Rust implementation anyway and we can look at the API design (pretty-formatted string representations? always move all namespace declarations to the root node when serializing a document?).

from delb-py.

the last cell in the docs/getting_started.ipynb also gives a great example where pretty is broken. so that demo-case could be taken as one test.

from delb-py.

i started looking into this which led me to realise that the serialization doesn't consider the xml:space attribute yet.

from delb-py.

after taking a few superficial glances, just some open ended threads:

• the _collapse_whitespace method implements the recommendations linked in Document.collapse_whitespace, iirc
• i assume that these recommendations generally make sense for text documents
• i also assume that remove_blank_text is rather harsh in its doings, i guess i would have used that option if it was compatible with the aforementioned recommendations
• does lxml's pretty serialization comply with these recommendations?
• a serialization must not add whitespace here, for example: la<hi rendition="u">la</hi>la
• can we agree that a pretty serialization must yield a result that is, when again parsed with the collapse_whitespace option, identical to the parsing of an unpretty Serialisat?

from delb-py.