I have seen the most problem in the PDF builds, but the HTML builds also
seem to do funny stuff with more vertical space if you don't run your
first and last text against the opening and closing <screen> tags,
respectively. In other words, given these two examples:
<!-- first example -->
<screen>
<computeroutput>
foo
bar
</computeroutput>
</screen>
<!-- second example -->
<screen><computeroutput>foo
bar</computeroutput></screen>
The second example renders into a more pleasing vertical context,
without a lot of wasted space. The PDF, IIRC, was particularly ugly if
you didn't use the second form, but I seem to remember the HTML also was
noticeably different.
I can't get anything to build PDF right now to test this, but in HTML I
see the same thing as output, so am not sure of the advantage of the
different way you suggest it.
I do notice that doing sgml-fill-paragraph will make them line up like:
<screen><computeroutput>foo bar</computeroutput></screen>
Then introducing the line break gets the functionality you mention.
Ultimately, I think we will want to drop the redundant <computeroutput>
and use a CDATA container instead. In that case, _all_ whitespace will
be considered for certain. All content will need to use the left margin
of the XML as starting point for indention.