3
0
Fork 0
forked from mirrors/nixpkgs
nixpkgs/nixos/lib/make-options-doc/postprocess-option-descriptions.xsl

116 lines
4.8 KiB
XML
Raw Normal View History

<?xml version="1.0"?>
<xsl:stylesheet version="1.0"
xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
xmlns:str="http://exslt.org/strings"
xmlns:exsl="http://exslt.org/common"
xmlns:db="http://docbook.org/ns/docbook"
xmlns:nixos="tag:nixos.org"
extension-element-prefixes="str exsl">
<xsl:output method='xml' encoding="UTF-8" />
<xsl:template match="@*|node()">
<xsl:copy>
<xsl:apply-templates select="@*|node()" />
</xsl:copy>
</xsl:template>
<xsl:template name="break-up-description">
<xsl:param name="input" />
<xsl:param name="buffer" />
<!-- Every time we have two newlines following each other, we want to
break it into </para><para>. -->
<xsl:variable name="parbreak" select="'&#xa;&#xa;'" />
<!-- Similar to "(head:tail) = input" in Haskell. -->
<xsl:variable name="head" select="$input[1]" />
<xsl:variable name="tail" select="$input[position() &gt; 1]" />
<xsl:choose>
<xsl:when test="$head/self::text() and contains($head, $parbreak)">
<!-- If the haystack provided to str:split() directly starts or
ends with $parbreak, it doesn't generate a <token/> for that,
so we are doing this here. -->
<xsl:variable name="splitted-raw">
<xsl:if test="starts-with($head, $parbreak)"><token /></xsl:if>
<xsl:for-each select="str:split($head, $parbreak)">
<token><xsl:value-of select="node()" /></token>
</xsl:for-each>
<!-- Something like ends-with($head, $parbreak), but there is
no ends-with() in XSLT, so we need to use substring(). -->
<xsl:if test="
substring($head, string-length($head) -
string-length($parbreak) + 1) = $parbreak
"><token /></xsl:if>
</xsl:variable>
<xsl:variable name="splitted"
select="exsl:node-set($splitted-raw)/token" />
<!-- The buffer we had so far didn't contain any text nodes that
contain a $parbreak, so we can put the buffer along with the
first token of $splitted into a para element. -->
<para xmlns="http://docbook.org/ns/docbook">
<xsl:apply-templates select="exsl:node-set($buffer)" />
<xsl:apply-templates select="$splitted[1]/node()" />
</para>
<!-- We have already emitted the first splitted result, so the
last result is going to be set as the new $buffer later
because its contents may not be directly followed up by a
$parbreak. -->
<xsl:for-each select="$splitted[position() &gt; 1
and position() &lt; last()]">
<para xmlns="http://docbook.org/ns/docbook">
<xsl:apply-templates select="node()" />
</para>
</xsl:for-each>
<xsl:call-template name="break-up-description">
<xsl:with-param name="input" select="$tail" />
<xsl:with-param name="buffer" select="$splitted[last()]/node()" />
</xsl:call-template>
</xsl:when>
<!-- Either non-text node or one without $parbreak, which we just
want to buffer and continue recursing. -->
<xsl:when test="$input">
<xsl:call-template name="break-up-description">
<xsl:with-param name="input" select="$tail" />
<!-- This essentially appends $head to $buffer. -->
<xsl:with-param name="buffer">
<xsl:if test="$buffer">
<xsl:for-each select="exsl:node-set($buffer)">
<xsl:apply-templates select="." />
</xsl:for-each>
</xsl:if>
<xsl:apply-templates select="$head" />
</xsl:with-param>
</xsl:call-template>
</xsl:when>
<!-- No more $input, just put the remaining $buffer in a para. -->
<xsl:otherwise>
<para xmlns="http://docbook.org/ns/docbook">
<xsl:apply-templates select="exsl:node-set($buffer)" />
</para>
</xsl:otherwise>
</xsl:choose>
</xsl:template>
<xsl:template match="nixos:option-description">
<xsl:choose>
<!--
Only process nodes that are comprised of a single <para/> element,
because if that's not the case the description already contains
</para><para> in between and we need no further processing.
-->
<xsl:when test="count(db:para) > 1">
<xsl:apply-templates select="node()" />
</xsl:when>
<xsl:otherwise>
<xsl:call-template name="break-up-description">
<xsl:with-param name="input"
select="exsl:node-set(db:para/node())" />
</xsl:call-template>
</xsl:otherwise>
</xsl:choose>
</xsl:template>
</xsl:stylesheet>