On Mon, Jul 14, 2014 at 09:11:05PM +0530, Rupinder Singh wrote: > On Sat, Jul 12, 2014 at 10:47:48PM +0100, NetSurf Browser Project wrote: > > > I don't understand this changeset. Why have you exposed the tokeniser's > > internal state to the outside world? It is extremely dangerous to do > > this, as it makes it utterly unclear what triggers a state transition. > > Additionally, only the tokeniser needs to know what state it is in; the > > client of the tokeniser must treat it as a black box. > > > > > I had the same concerns. Wanted to, but couldn't discuss it earlier. I > did it as the specs say that the treebuilder must change tokeniser's state > here, under "script" start tag, step 7. > http://www.whatwg.org/specs/web-apps/current-work/multipage/tree-construction.html#parsing-main-inhead > .. There can be workarounds to avoid this. Please suggest some, or I'll try > to work out some.
Well, the existing precedent for this kind of thing is the tokeniser's content model, or the pause flag. Essentially, hubbub_tokeniser_setopt is used to inject configuration into the tokeniser, which it is perfectly at liberty to ignore, if it wants. The point here is that the injection of configuration is a *request* to do something, which allows the internals of the tokeniser to be kept private from its client. J.
