Major changes for WeasyPrint

[liZe]

No description provided.

[liZe]

Main changes:

• ElementWrapper doesn't try to mimic ElementTree's Element, Element's attributes and methods (get) are not available from the wrapper,
• a base_url attribute has been added to the wrapper and is automatically taken from the parent if needed,
• match returns (specificity, order, pseudo, payload) instead of just the payload.

Fixes (both issues found by Acid2):

• selectors with pseudo-classes missing the universal selectors (like html :hover) are now parsed correctly,
• the declaration order is explicitely kept to fix cases where two selectors have the same global specificity but their last simple selectors don't have the same specificity:

<style>
  p .class { color: red }
  .class p { color: blue }
</style>
<p class="class">
  <p class="class">
    This text was red and is now blue.
  </p>
</p>

I'm not really happy with this API, especially with what's returned by match. The goal of this PR is to have a 0.1 version released quickly, use it in WeasyPrint and have a discussion with other users interested in cssselect2 (about the API, #1 and everything else) with a real-life working example.

Selectors level 4 (in cssselect2) and Named pages (in WeasyPrint) are some of the next challenges 😉.

@SimonSapin is that OK for you?

[SimonSapin]

Feel free to do whatever’s good for WeasyPrint with this repo :) Maybe even move into the WeasyPrint repository? You could still publish packages to PyPI separately if desired. (BTW I just gave you PyPI access.)

I haven’t looked at the diffs, but based on your "main changes" description:

• I think ElementWrapper should go further and not be a wrapper at all. Maybe eliminate the children list and instead have previous_siblings, next_sibling, first_child, and last_child links to other elements directly, if that makes matching easier.
• Unless you want to support xml:base (which is fairly obscure), the base URL is the same for an entire documents. Maybe it doesn’t need to be stored on each element?
• We’ve discussed the match return type on IRC, I think this change should not be needed if you have multiple Matchers. But whatever you find convenient.
• I think declaration order was preserved by using a stable sort. Was that buggy?

[liZe]

Maybe even move into the WeasyPrint repository? You could still publish packages to PyPI separately if desired. (BTW I just gave you PyPI access.)

I'd like to see if anyone else is interested in cssselect2 before merging it into WeasyPrint.

I think ElementWrapper should go further and not be a wrapper at all. Maybe eliminate the children list and instead have previous_siblings, next_sibling, first_child, and last_child links to other elements directly, if that makes matching easier.

I agree, I'll do that later.

Unless you want to support xml:base (which is fairly obscure), the base URL is the same for an entire documents. Maybe it doesn’t need to be stored on each element?

Yes, I've added it because it was easier, I should remove it before the release (I think that the hard part is already done in WeasyPrint).

We’ve discussed the match return type on IRC, I think this change should not be needed if you have multiple Matchers. But whatever you find convenient.

I think that having one matcher helps to keep the code simple in WeasyPrint for now.

I think declaration order was preserved by using a stable sort. Was that buggy?

Yes, because in match the relevant_selectors are added in an order related to the specificity of their last simple selector.