Converting XML to plain text - how should I ignore/handle whitespace in the XSLT?

There are three reasons for getting unwanted whitespace in the result of an XSLT transformation:

1. whitespace that comes from between nodes in the source document
2. whitespace that comes from within nodes in the source document
3. whitespace that comes from the stylesheet

I'm going to talk about all three because it can be hard to tell where whitespace comes from so you might need to use several strategies.

To address the whitespace that is between nodes in your source document, you should use <xsl:strip-space> to strip out any whitespace that appears between two nodes, and then use <xsl:preserve-space> to preserve the significant whitespace that might appear within mixed content. For example, if your source document looks like:

<ul>  <li>This is an <strong>important</strong> <em>point</em></li></ul>

then you will want to ignore the whitespace between the <ul> and the <li> and between the </li> and the </ul>, which is not significant, but preserve the whitespace between the <strong> and <em> elements, which is significant (otherwise you'd get "This is an **important***point*"). To do this use

<xsl:strip-space elements="*" /><xsl:preserve-space elements="li" />

The elements attribute on <xsl:preserve-space> should basically list all the elements in your document that have mixed content.

Aside: using <xsl:strip-space> also reduces the size of the source tree in memory, and makes your stylesheet more efficient, so it's worth doing even if you don't have whitespace problems of this sort.

To address the whitespace that appears within nodes in your source document, you should use normalize-space(). For example, if you have:

<dt>  a definition</dt>

and you can be sure that the <dt> element won't hold any elements that you want to do something with, then you can do:

<xsl:template match="dt">  ...  <xsl:value-of select="normalize-space(.)" />  ...</xsl:template>

The leading and trailing whitespace will be stripped from the value of the <dt> element and you will just get the string "a definition".

To address whitespace coming from the stylesheet, which is perhaps the one you're experiencing, is when you have text within a template like this:

<xsl:template match="name">  Name:  <xsl:value-of select="." /></xsl:template>

XSLT stylesheets are parsed in the same way as the source documents that they process, so the above XSLT is interpreted as a tree that holds an <xsl:template> element with a match attribute whose first child is a text node and whose second child is a <xsl:value-of> element with a select attribute. The text node has leading and trailing whitespace (including line breaks); since it's literal text in the stylesheet, it gets literally copied over into the result, with all the leading and trailing whitespace.

But some whitespace in XSLT stylesheets get stripped automatically, namely those between nodes. You don't get a line break in your result because there's a line break between the <xsl:value-of> and the close of the <xsl:template>.

To get only the text you want in the result, use the <xsl:text> element like this:

<xsl:template match="name">  <xsl:text>Name: </xsl:text>  <xsl:value-of select="." /></xsl:template>

The XSLT processor will ignore the line breaks and indentation that appear between nodes, and only output the text within the <xsl:text> element.

Are you using indent="no" in your output tag?

<xsl:output method="text" indent="no" />

Also if you're using xsl:value-of you can use the disable-output-escaping="yes" to help with some whitespace issues.

@JeniT's answer is great, I just want to point out a trick for managing whitespace. I'm not certain it's the best way (or even a good way), but it works for me for now.

("s" for space, "e" for empty, "n" for newline.)

<?xml version="1.0" encoding="UTF-8"?><!DOCTYPE xsl:transform [  <!ENTITY s "<xsl:text xmlns:xsl='http://www.w3.org/1999/XSL/Transform'> </xsl:text>" >  <!ENTITY s2 "<xsl:text xmlns:xsl='http://www.w3.org/1999/XSL/Transform'>  </xsl:text>" >  <!ENTITY s4 "<xsl:text xmlns:xsl='http://www.w3.org/1999/XSL/Transform'>    </xsl:text>" >  <!ENTITY s6 "<xsl:text xmlns:xsl='http://www.w3.org/1999/XSL/Transform'>      </xsl:text>" >  <!ENTITY e "<xsl:text xmlns:xsl='http://www.w3.org/1999/XSL/Transform'></xsl:text>" >  <!ENTITY n "<xsl:text xmlns:xsl='http://www.w3.org/1999/XSL/Transform'></xsl:text>" >]><xsl:transform version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform" xmlns:xs="http://www.w3.org/2001/XMLSchema"><xsl:output method="text"/><xsl:template match="/">  &e;Flush left, despite the indentation.&n;  &e;  This line will be output indented two spaces.&n;      <!-- the blank lines above/below won't be output -->  <xsl:for-each select="//foo">    &e;  Starts with two blanks: <xsl:value-of select="@bar"/>.&n;    &e;  <xsl:value-of select="@baz"/> The 'e' trick won't work here.&n;    &s2;<xsl:value-of select="@baz"/> Use s2 instead.&n;    &s2;    <xsl:value-of select="@abc"/>    <xsl:value-of select="@xyz"/>&n;    &s2;    <xsl:value-of select="@abc"/>&s;<xsl:value-of select="@xyz"/>&n;  </xsl:for-each></xsl:template></xsl:transform>

Applied to:

<?xml version="1.0" encoding="UTF-8"?><foo bar="bar" baz="baz" abc="abc" xyz="xyz"></foo>

Outputs:

Flush left, despite the indentation.  This line will be output indented two spaces.  Starts with two blanks: bar.baz The 'e' trick won't work here.  baz Use s2 instead.  abcxyz  abc xyz

The 'e' trick works prior to a text node containing at least one non-whitespace character because it expands to this:

<xsl:template match="/">  <xsl:text></xsl:text>Flush left, despite the indentation.<xsl:text></xsl:text>

Since the rules for stripping whitespace say that whitespace-only text nodes get stripped, the newline and indentation between the <xsl:template> and <xsl:text> get stripped (good). Since the rules say a text node with at least one whitespace character is preserved, the implicit text node containing " This line will be output indented two spaces." keeps its leading whitespace (but I guess this also depends on the settings for strip/preserve/normalize). The "&n;" at the end of the line inserts a newline, but it also ensures that any following whitespace is ignored, because it appears between two nodes.

The trouble I have is when I want to output an indented line that begins with an <xsl:value-of>. In that case, the "&e;" won't help, because the indentation whitespace isn't "attached" to any non-whitespace characters. So for those cases, I use "&s2;" or "&s4;", depending on how much indentation I want.

It's an ugly hack I'm sure, but at least I don't have the verbose "<xsl:text>" tags littering my XSLT, and at least I can still indent the XSLT itself so it's legible. I feel like I'm abusing XSLT for something it was not designed for (text processing) and this is the best I can do.

Edit:In response to comments, this is what it looks like without the "macros":

<xsl:template match="/">  <xsl:text>Flush left, despite the indentation.</xsl:text>  <xsl:text>  This line will be output indented two spaces.</xsl:text>  <xsl:for-each select="//foo">    <xsl:text>  Starts with two blanks: </xsl:text><xsl:value-of select="@bar"/>.<xsl:text></xsl:text>    <xsl:text>    </xsl:text><xsl:value-of select="@abc"/><xsl:text> </xsl:text><xsl:value-of select="@xyz"/><xsl:text></xsl:text>  </xsl:for-each></xsl:template>

I think that makes it less clear to see the intended output indentation, and it screws up the indentation of the XSL itself because the </xsl:text> end tags have to appear at column 1 of the XSL file (otherwise you get undesired whitespace in the output file).