Re: white-space patch

On Sun, Oct 02, 2011 at 11:09:25AM +0200, Johannes Hofmann wrote:
> On Sat, Oct 01, 2011 at 03:17:09PM -0300, Jorge Arellano Cid wrote:
> > On Sat, Oct 01, 2011 at 04:45:52PM +0000, corvid wrote:
> > > Jorge wrote:
> > > > On Sat, Oct 01, 2011 at 11:03:18AM -0300, Jorge Arellano Cid wrote:
> > > > > On Fri, Sep 30, 2011 at 08:54:39AM +0200, Johannes Hofmann wrote:
> > > > > > On Wed, Sep 28, 2011 at 05:49:10PM -0300, Jorge Arellano Cid wrote:
> > > > 
> > > > > > >   [...]
> > > > > > >   Please explain the non-deterministical behaviour at google.
> > > > > > 
> > > > > > If I search for e.g. dillo on google, the first line (the link) of a
> > > > > > result entry is sometimes broken into two lines, sometimes not.
> > > > > 
> > > > >   Thanks for the explanation.
> > > > > 
> > > > >   After some time reviewing it and guessing, I succeeded to isolate
> > > > > a small testcase which I'm reviewing now.
> > > > 
> > > >   Sorry, it was not a reliable testcase...  :-P
> > > 
> > > I haven't looked into this, but do you suppose it could have to do with the
> > > part in html.cc that goes
> > > 
> > >       if (isspace(buf[buf_index])) {
> > >          /* whitespace: group all available whitespace */
> > >          while (++buf_index < bufsize && isspace(buf[buf_index])) ;
> > >          Html_process_space(html, buf + token_start, buf_index - token_start);
> > >          token_start = buf_index;
> > >       } ...
> > > 
> > > that is, sends off whatever it's accumulated at the end of a packet?
> > 
> >   Not sure, but I don't think so.
> > 
> >   Without  the  whitespace  patch, dillo also doesn't seem to get
> > right the available horizontal space for the link [1].
> > 
> >   To  me it looks related to the way ParMin propagates upwards in
> > a textblock, but I'm just starting to look into it.
> 
> Yes, I also think it's related to that.
> Textblock::getExtremesImpl() starts from wrapRef and uses the
> precomputed line->maxWordMin, line->maxParMax, line->parMin, and
> line->parMax. From there it does compute the stuff on it's own.
> 
> So if the computation of line->maxWordMin, line->maxParMax,
> line->parMin, and line->parMax in wordWrap() doesn't agree with what
> Textblock::getExtremesImpl() does afterwards, the result becomes
> unreliable.

  FWIW, this old comment in the code (not present today)
may be a good hint of what happens now:

  textblock.cc, Textblock::wordWrap()

   /* NOTE: Most code relies on that all values of nowrap are equal for all
    * words within one line. */

-- 
  Cheers
  Jorge.-