[imps] Missing tree-construction tests (after Twintsam code-coverage analysis)

[Thomas Broyer]

Still replying to myself...

>  >  If someone wants (and have time) to scan through all the tests to see
>  >  if they're still OK wrt the current draft; I suggest reorganizing
>  >  tests (e.g. grouped by insertion mode) and moving them into
>  >  http://html5.googlecode.com at the same time.
>
>  I've start such a work, but re-creating all the tests from scratch.
>  Read the commit log for explanations (it should probably be moved into
>  a README):
>  http://code.google.com/p/html5/source/detail?r=145
>
>  I've no yet used those new files with Twintsam or html5lib, I just ran
>  each individual test through html5lib's parse.py (with -e and --tree
>  options) to check the output against my understanding of the spec.

I ran the tests through html5lib. There are a few things I changed in
the test-file format which required some changes to html5lib:
 - the DOCTYPEs are serialized with their PUBLIC and SYSTEM
identifiers if they are present (and non-empty, because html5lib
doesn't tell the difference). I might add a "#compatibility-mode"
sectio with values "quirks", "no quirks" and "limited quirks" (and
move the tests into a compatibility-mode.dat file)
 - the comments are serialized without extra spaces (the current test
format expects "<!-- comment -->" if the comment data is "comment", my
new tests expect "<!--comment-->")

To make it easier to run the tests in html5lib, I'm thinking about
putting a .patch file into html5lib's SVN repository: apply the patch
and run tests.

What do you think?

Some of the tests revealed wrong after analyzing why html5lib didn't
pass them (cut/paste is really a bad practice ;-) ).

-- 
Thomas Broyer