generated html can be added many time in one page relying on line number
to collapse is dangerous.

Walking the dom to  get next children is more reliable

in particular tings like 'code' are conflicting when embeding the page

will allow refactor of the cython magic not to parse temporary files
and not to leak css into IPython notebooks

this should help to decouple from the filesystem and heuristic in
filenames

utility code construction for struct conversion requires constructing the field conversion functions as well

--HG--
extra : amend_source : af016229b6c8c40168070e4d3a9581ed12500d3f

utility code construction for struct conversion requires constructing the field conversion functions as well

--HG--
extra : transplant_source : %A3%81%87%E2%0A%20%DA%7D%7E%5B%A0%9B%80i%F9%8Az%9B%F3%DC

Compute each Makefile variable immediately

Otherwise the variables get re-computed over and over again every time
they are used, slowing things up.  Especially if someone adapts the
Makefile to several targets in their own real project. :)

revert to using codecs.open() for output files instead of io.open() as the latter cannot write str objects in text mode under Py2 (and we erroneously do that in some spots)

make encoding detection code a bit faster if the first two source file lines are longer than 250 bytes