On Fri, 16 Dec 2005 11:40 pm, Adam Strzelecki wrote:
> Hi Jeremias,
>
> > All test cases pass, so it's a special situation. Please send a
> > test file. Thanks.
>
> Attached.
>
> Seems that I'm all of special situations ;)
It is not a special situation at all. Simply having a in your
text is enough to trigger this regression. And yes, it is an regression
caused by the "fix" for the disappearing empty line problem:
<fo:block> </fo:block>
We don't have test cases in our test suite for non breaking spaces and
therefore this regression wasn't caught.
After some debugging it is obvious to me what is happening. However, it
is much less obvious how to fix it. In brief:
The fix added a zero width Knuth box to the element list created for a
nbsp. This was done for two reasons:
a) to avoid it being removed by the removeElementsForTrailingSpaces()
method. That method is wrong in the first place as it knowingly tries
to remove non breaking spaces which IMO should never be removed.
b) However, changing that method to not remove nbsp would give us Knuth
paragraphs containing only Glue and Penalty elements. This then causes
our implementation of the Knuth algorithm to 'spit the dummy'. Luca,
why does our line breaking algorithm insist on having at least one Box
in a paragraph? Is that inherent in the Knuth algorithm, i.e. can't it
deal with empty paragraphs, that is paragraphs containing only Glue/Pen
elements?
However, this now causes consecutive Knuth boxes being created when the
nbsp is for example between two words. This then triggers the
addALetterSpaceMethod but it fails because the Knuth sequences are not
any more like what it expects.
It is well documented and acknowledged in other threads on this mailing
list that the design of this area of the code (generating of the Knuth
sequences for inline's) is flawed and this is just another
manifestation of the inherent design problems here.
I'll see if I can find another hack to workaround this particular
problem but I don't feel comfortable with the whole scenario.
>
> BTW. Maybe it is worth to put some special message at
> src\java\org\apache\fop\cli\Main.java line 185
>
> options.getLogger().error("This is probably bug of FOP version blah
> bla... contact us at blah blah or send bug report at blah blah.", e);
> e.printStackTrace();
>
> instead options.getLogger().error("Exception", e);
>
> .. coz I needed to modify Main.java in order to see the stack trace.
>
> Cheers,
Manuel
