[xsl] broken text surrounding an entity I want to drop?

Subject: [xsl] broken text surrounding an entity I want to drop?
From: "Trevor Nicholls" <trevor@xxxxxxxxxxxxxxxxxx>
Date: Tue, 13 Sep 2005 14:50:55 +1200

My source XML file contains a myriad of <A id=something></A> entities which
for the most part I wish to drop. I am using an identity template plus the
following to do this:

<!-- drop A tags which have no content -->
<xsl:template match="A">
<xsl:if test="* or text() or string(.)">
<xsl:apply-templates select="@*|node()"/>

Unfortunately and unsurprisingly this is too naove; it drops the <a> tags OK
but leaves me with broken text. If I look at the input source in a text
editor I can see that these tags are placed (arbitrarily so far as I can
a. between newlines
b. after a newline and before text
c. after text and before a newline
d. between text strings

In cases a and b my output preserves the newlines and my later transforms
which normalize whitespace are fine.
In cases c and d the tag may be in the middle of a word or at either end of
a word. When at either end it is again not a problem as the output XML
contains a newline which is normalized acceptably. My difficulty comes with
this kind of input:

<A ID="something"></A>
To delete a n<A ID="something"></A>

Note that there are several thousand node types which can potentially hold
this kind of text content, so writing a "Body" template to manage it isn't
really feasible.
Ideally this input should become
<Body>To delete a node:</Body>

But of course the transform I'm using isn't doing this, I'm getting
<Body>To delete a n

My ham-fisted attempts to come up with templates which
(a) *reliably* identify this situation, and
(b) *don't* lead to my dropping huge screeds of wanted XML
are failing miserably. This doesn't seem like a terribly unusual
requirement, but I can't find an answer in the FAQ or my current set of
books. I've also read the (otherwise helpful) "controlling whitespace"
articles by Bob DuCharme on xml.com.

Could somebody please point me towards the right technique to use here?


Current Thread