% language=uk
% \startluacode
% xml.trace_lpath = true
% \stopluacode
\startcomponent mk-xml
\environment mk-environment
\chapter{XML revisioned}
{\em The code dealing with \XML\ is evolving and the following
text might be outdated. So, in case of doubt, check the manual.}
\subject{the parser}
For quite a while \CONTEXT\ has built-in support for \XML\ processing and
at \PRAGMA\ we use this extensively. One of the first things I tried to deal
with in \LUA\ was \XML, and now that we have \LUATEX\ up and running it's
time to investigate this a bit more. First we'll have a look at the basic
functions, the \LUA\ side of the game.
We load an \XML\ file as follows (the \type {document} namespace
is predefined in \CONTEXT):
\startbuffer
\startluacode
document.xml = document.xml or { } -- define namespace
document.xml = xml.load("mk-xml.xml") -- load the file
\stopluacode
\stopbuffer
\typebuffer \getbuffer
The loader constructs a table representing the document structure, including
whitespace, so let's serialize the code and see what shows up:
\startbuffer
\startluacode
local prn = xml.newhandlers { handle = tex.sprint }
tex.sprint("\\starttyping")
xml.serialize(document.xml, prn)
tex.sprint("\\stoptyping")
\stopluacode
\stopbuffer
\typebuffer
In the first version of the serializer, we could pass extra function
arguments that controlled the way content was processed. This method
has now been replaced by handlers. In this example we create a
simple handler where the \type {handle} function is responsible
for the final print.
\getbuffer
This already gives us a rather basic way to manipulate documents and
this method is even not that slow because we bypass \TEX\ reading from
file.
\startbuffer
\startluacode
local str = " hello world "
local prn = xml.newhandlers { handle = tex.sprint }
tex.sprint("\\starttyping")
xml.serialize(xml.convert(str),prn)
tex.sprint("\\stoptyping")
\stopluacode
\stopbuffer
\typebuffer
Watch the extra print argument, we need this because otherwise the
verbatim mode will not work out well.
\getbuffer
You need to keep in mind that in these examples we print to \TEX\ under
the current catcode regime.
You can save a \XML\ table with the command:
\starttyping
\startluacode
xml.save(document.xml,"newfile.xml")
\stopluacode
\stoptyping
These examples show that you have access to \XML\ files from
within your document. If you want to convert the table to just a
string, you can use \type {xml.tostring}. Actually, this method is
automatically used for occasions where \LUA\ wants to print an
\XML\ table or wants to join string snippets. However, as we are
inside \TEX, we need to print to \TEX\ instead of the console or
file. For this we use specialized handlers.
The reason why I wrote the \XML\ parser is that we need it in the
utilities (so it has to provide access to the content of elements)
as well as in the text processing (so it needs to provide some
manipulation features). To serve both we have implemented a subset
of what standard \XML\ tools qualify as path based searching.
\startbuffer
\startluacode
xml.sprint(xml.first(document.xml, "/one/three/some"))
\stopluacode
\stopbuffer
\typebuffer
The result of this snippet is the content of the first element
that matches the specification: \quote{\getbuffer}. As you can
see, this comes out rather verbose. The reason for this is that we
need to enter \XML\ mode in order to get such a snippet
interpreted.
Below we give a few more variants, this time
we use a generic filter:
\startbuffer
\startluacode
xml.sprint(xml.filter(document.xml, "/one/three/some"))
\stopluacode
\stopbuffer
\typebuffer result: \astype{\getbuffer}
\startbuffer
\startluacode
xml.sprint(xml.filter(document.xml, "/one/three/some/first()"))
\stopluacode
\stopbuffer
\typebuffer result: \astype{\getbuffer}
\startbuffer
\startluacode
xml.sprint(xml.filter(document.xml, "/one/three/some[1]"))
\stopluacode
\stopbuffer
\typebuffer result: \astype{\getbuffer}
\startbuffer
\startluacode
xml.sprint(xml.filter(document.xml, "/one/three/some[-1]"))
\stopluacode
\stopbuffer
\typebuffer result: \astype{\getbuffer}
\startbuffer
\startluacode
xml.sprint(xml.filter(document.xml, "/one/three/some/texts()"))
\stopluacode
\stopbuffer
\typebuffer result: \astype{\getbuffer}
\startbuffer
\startluacode
xml.sprint(xml.filter(document.xml, "/one/three/some[2]/text()"))
\stopluacode
\stopbuffer
\typebuffer result: \astype{\getbuffer}
The next lines shows some more variants. There are more than these and
we will extend the repertoire over time. If needed you can define
additional handlers.
\subject{performance}
Before we continue with more examples, a few remarks about the
performance. The first version of the parser was an enhanced
version of the one presented in the \LUA\ book: support for
namespaces, processing instructions, comments, cdata and doctype,
remapping and a few more things. When playing with the parser I
was quite satisfied about the performance. However, when I started
experimenting with 40~megabyte files, the preprocessing (needed
for the special elements) started to become more noticeable. For
smaller files its 40\% overhead is not that disturbing, but for
large files \unknown\
The current version uses \LPEG. We follow the same approach as
before, stack and top and such but this time parsing is about
twice as fast which is mostly due to the fact that we don't have
to prepare the stream for cdata, doctype etc. Loading the
mentioned large file took 12.5 seconds (1.5 for file io and the
rest for tree building) on my laptop (a 2.3 Ghz Core Duo running
Windows Vista). With the \LPEG\ implementation we got that down to
less 7.3 seconds. Loading the 14 interface definition files (2.6
meg) went down from 1.05 seconds to 0.55 seconds. Namespace
related issues take some 10\% of this.
Of course these numbers might change over time. For instance, we
now have the second implementation of the filter mechanism which
is more advanced and maybe somewhat slower on some tasks.
\subject{patterns}
We will not implement complete \XPATH\ functionality, but only the
features that make sense for documents that are well structured
and needs to be typeset. In addition we (will) implement text
manipulation functions. Of course speed is also a consideration
when implementing such mechanisms.
The following list is not complete (after all here we only give an
impression of the development) but it gives a good impression.
\nonknuthmode
\starttabulate[|l|c|l|]
\NC \bf pattern \NC \bf supported \NC \bf comment \NC \NR
\HL
\NC \type{a} \NC \star \NC not anchored \NC \NR
\NC \type{!a} \NC \star \NC not anchored,negated \NC \NR
\NC \type{a/b} \NC \star \NC anchored on preceding \NC \NR
\NC \type{/a/b} \NC \star \NC anchored (current root) \NC \NR
\NC \type{^a/c} \NC \star \NC anchored (current root) \NC \NR
\NC \type{^^/a/c} \NC todo \NC anchored (document root) \NC \NR
\NC \type{a/*/b} \NC \star \NC one wildcard \NC \NR
\NC \type{a//b} \NC \star \NC many wildcards \NC \NR
\NC \type{a/**/b} \NC \star \NC many wildcards \NC \NR
\NC \type{.} \NC \star \NC ignored self \NC \NR
\NC \type{..} \NC \star \NC parent \NC \NR
\NC \type{a[5]} \NC \star \NC index upwards \NC \NR
\NC \type{a[-5]} \NC \star \NC index downwards \NC \NR
\NC \type{a[position()=5]} \NC maybe \NC \NC \NR
\NC \type{a[first()]} \NC maybe \NC \NC \NR
\NC \type{a[last()]} \NC maybe \NC \NC \NR
\NC \type{(b|c|d)} \NC \star \NC alternates (one of) \NC \NR
\NC \type{b|c|d} \NC \star \NC alternates (one of) \NC \NR
\NC \type{!(b|c|d)} \NC \star \NC not one of \NC \NR
\NC \type{a/(b|c|d)/e/f} \NC \star \NC anchored alternates \NC \NR
\NC \type{(c/d|e)} \NC not likely \NC nested subpaths \NC \NR
\NC \type{a/b[@bla]} \NC \star \NC any value of \NC \NR
\NC \type{a/b/@bla} \NC \star \NC any value of \NC \NR
\NC \type{a/b[@bla='oeps']} \NC \star \NC equals value \NC \NR
\NC \type{a/b[@bla=='oeps']} \NC \star \NC equals value \NC \NR
\NC \type{a/b[@bla<>'oeps']} \NC \star \NC different value \NC \NR
\NC \type{a/b[@bla!='oeps']} \NC \star \NC different value \NC \NR
\TB
\NC \type{...../attribute(id)} \NC \star \NC \NC \NR
\NC \type{...../attributes()} \NC \star \NC \NC \NR
\NC \type{...../text()} \NC \star \NC \NC \NR
\NC \type{...../texts()} \NC \star \NC \NC \NR
\NC \type{...../first()} \NC \star \NC \NC \NR
\NC \type{...../last()} \NC \star \NC \NC \NR
\NC \type{...../index(n)} \NC \star \NC \NC \NR
\NC \type{...../position(n)} \NC \star \NC \NC \NR
\TB
\NC \type{root::} \NC \star \NC \NC \NR
\NC \type{parent::} \NC \star \NC \NC \NR
\NC \type{child::} \NC \star \NC \NC \NR
\NC \type{ancestor::} \NC \star \NC \NC \NR
\NC \type{preceding-sibling::} \NC not soon \NC \NC \NR
\NC \type{following-sibling::} \NC not soon \NC \NC \NR
\NC \type{preceding-sibling-of-self::} \NC not soon \NC \NC \NR
\NC \type{following-sibling-or-self::} \NC not soon \NC \NC \NR
\NC \type{descendent::} \NC \star \NC \NC \NR
\NC \type{descendent-or-self::} \NC \star \NC \NC \NR
\NC \type{preceding::} \NC not soon \NC \NC \NR
\NC \type{following::} \NC not soon \NC \NC \NR
\NC \type{self::node()} \NC not soon \NC \NC \NR
\NC \type{id("tag")} \NC not soon \NC \NC \NR
\NC \type{node()} \NC not soon \NC \NC \NR
\stoptabulate
This list shows that it is also possible to ask for more matches at
once. Namespaces are supported (including a wildcard) and there are
mechanisms for namespace remapping.
\startbuffer
\startluacode
lxml.concat(document.xml,"/one/(three|five)/some",", "," and ")
\stopluacode
\stopbuffer
\typebuffer
We get: \astype{\getbuffer} and if we say:
\startbuffer
\startluacode
lxml.concat(document.xml,"/one/(three|five)/some",", "," and ",
true)
\stopluacode
\stopbuffer
\typebuffer
We get: \quote {\getbuffer}.
Watch how we use the \type {lxml} namespace here! Here live the
functions that pipe the result to \TEX.
\startbuffer
\startluacode
lxml.count(document.xml,"/one/(three|five)/some")
\stopluacode
\stopbuffer
There a several helper functions, like \type {xml.count} which in this case
returns~\getbuffer.
\typebuffer
Functions like this gives the opportunity to loop over lists of elements
by index.
\subject{manipulations}
We can manipulate elements too. The next code will add some elements
at specific locations.
\startbuffer
\startluacode
xml.before(document.xml,"xml:///one/three/some","ok")
xml.after (document.xml,"xml:///one/three/some","ok")
tex.sprint("\\starttyping")
xml.sprint(lxml.filter(document.xml,"/one/three"))
tex.sprint("\\stoptyping")
\stopluacode
\stopbuffer
\typebuffer
And indeed, we suddenly have a couple of \quote {ok}'s there:
\getbuffer
Of course wel can also delete elements:
\startbuffer
\startluacode
xml.delete(document.xml,"/one/three/some")
xml.delete(document.xml,"/one/three/af")
tex.sprint("\\starttyping")
xml.sprint(lxml.filter(document.xml,"/one/three"))
tex.sprint("\\stoptyping")
\stopluacode
\stopbuffer
\typebuffer
Now we have:
\getbuffer
Replacing an element is also possible. The replacement can be a
table (representing elements) or a string which is then converted
into a table first.
\startbuffer
\startluacode
xml.replace(document.xml,"/one/three/be","done")
tex.sprint("\\starttyping")
xml.sprint(lxml.filter(document.xml,"/one/three"))
tex.sprint("\\stoptyping")
\stopluacode
\stopbuffer
\typebuffer
And indeed we get:
\getbuffer
These are just a few features of the library. I will add some more (rather) generic
manipulaters and extend the functionality of the existing ones. Also, there will
be a few manipulation functions that come in handy when preparing texts for
processing with \TEX\ (most of the \XML\ that I deal with is rather dirty and needs
some cleanup).
\subject{streaming trees}
Eventually we will provies series of convenient macros that will provide an
alternative for most of the \MKII\ code. In \MKII\ we have a streaming parser, which
boils down to attaching macros to elements. This includes a mechanism for saving
an restoring data, but this is not always convenient because one also has to
intercept elements that needs to be hidden.
In \MKIV\ we do things different. First we load the complete document in memory (a
\LUA\ table). Then we flush the elements that we want to process. We can associate
setups with elements using the filters mentioned before. We can either use \TEX\ or
use \LUA\ to manipulate content. Instead if a streaming parser we now have a mixture
of streaming and tree manipulation available. Interesting is that the \XML\ loader
is pretty fast and piping data to \TEX\ is also efficient. Since we no longer need to
manipulate the elements in \TEX\ we gain processing time too, so in practice we have
now much faster \XML\ processing available.
To give you an idea we show a few commands:
\startbuffer
\xmlload {main}{mk-xml.xml}
\stopbuffer
\typebuffer \getbuffer
So that we can do things like (there are and will be a few more):
\starttabulate[|l|l|l|]
\NC \bf command \NC \bf arguments \NC \bf result \NC \NR
\NC \type {\xmlfirst} \NC \type {{main} {/one/three/some}} \NC \xmlfirst{main}{/one/three/some} \NC \NR
\NC \type {\xmllast } \NC \type {{main} {/one/three/some}} \NC \xmllast {main}{/one/three/some} \NC \NR
\NC \type {\xmlindex} \NC \type {{main} {/one/three/some} {2}} \NC \xmlindex{main}{/one/three/some}{2} \NC \NR
\stoptabulate
There is a set of about 30 commands that operates on the tree: loading, flushing,
filtering, associating setups and code in modules to elements. For instance when
one uses so called cals||tables, the processing is automatically activates when the
namespace can be resolved. Processing is collected in setups and those registered
are these are processed after loading the tree. In the following example we register
a handler for content that needs to end up bold.
\starttyping
\startxmlsetups xml:mysetups
\xmlsetsetup{\xmldocument}{bold|bf}{xml:handlebold}
\stopxmlsetups
\xmlregistersetup{xml:mysetups}
\startxmlsetups xml:handlebold
\dontleavehmode
\bgroup
\bf
\xmlflush{#1}
\egroup
\stopxmlsetups
\stoptyping
In this example \type {#1} represents the root of the subtree. Say that we
want to process an index entry which is coded as follows:
\starttyping
whatever
whatever
\stoptyping
We register an additional handler (here the \type {*} is a shortcut for
using the element's tag as setup name):
\starttyping
\startxmlsetups xml:mysetups
\xmlsetsetup{\xmldocument}{bold|bf}{xml:handlebold}
\xmlsetsetup{\xmldocument}{index}{*}
\stopxmlsetups
\xmlregistersetup{xml:mysetups}
\startxmlsetups index
\index[\xmlfirst{#1}{key}]{\xmlfirst{#1}{entry}}
\stopxmlsetups
\stoptyping
In practice \MKIV\ definitions are more compact than the comparable
\MKII\ ones, especially for more complex constructs (tables and such).
\starttyping
\defineXMLenvironment
[index]
{\bgroup
\defineXMLsave[key]%
\defineXMLsave[entry]}
{\index[\XMLflush{key}]{\XMLflush{entry}}%
\egroup}
\stoptyping
This looks compact, but keep in mind that we also need to get rid of
spurry spaces and when the code grows, we usually use setups to separate
the definition from the code. In any case, the \MKII\ solution involves
a few definitions as well as saving the content of elements. This is often
much more costly than the \MKIV\ method where we only locate and flush
content. Of course the document is stored in memory, but that happens
pretty fast: storing the 14~files (2~per interface) that define the \CONTEXT\
user interface takes .85 seconds on a 2.3 Ghz Core Duo (Windows Vista) which
is not that bad if you take into account that we're talking of 2.7 megabytes
of highly structured data (many elements and attributes, not that much text).
Loading one of these files using \MKII\ code (for storing elements) takes
many more seconds.
I didn't do extensive speed tests yet but for normal streamed
processing of simple documents the penalty of loading the tree can be
neglected. When comparing traditional \MKII\ code like:
\starttyping
\defineXMLargument [title][id=] {\subject[\XMLop{at}]}
\defineXMLenvironment[p] {} {\par}
\starttext
\processXMLfilegrouped{testspeed.xml}
\stoptext
\stoptyping
with its \MKIV\ counterpart:
\starttyping
\startxmlsetups document
\xmlsetsetup\xmldocument{title|p}{*}
\stopxmlsetups
\xmlregistersetup{document}
\startxmlsetups title
\section[\xmlatt{#1}{id}]{\xmlcontent{#1}{/}}
\stopxmlsetups
\startxmlsetups p
\xmlflush{#1}\endgraf
\stopxmlsetups
\starttext
\processXMLfilegrouped{testspeed.xml}
\stoptext
I found that processing a one megabyte file with some 400 sections
takes the same runtime for both approaches. However, as soon as more
complex manipulations enter the game the \MKIV\ method starts taking
less time. Think of the manipulations needed for \MATHML\ or converting
tables into something that \CONTEXT\ can handle. Also, when we deal
with documents where we need to ignore large portions of shuffle content
around, the traditional method also has to store data in memory and in
that case \MKII\ code always loses from \MKIV\ code. Of course any speed
we gain in handling \XML\ is lost on processing complex fonts and
attributes but there we gain in quality.
\stoptyping
Another advantage of the \MKIV\ mechanisms is that we suddenly have so called
fully expandable \XML\ handling. All manipulations take place in \LUA\ and
there is no interfering code at the \TEX\ end.
\subject{examples}
For the path freaks we now show what patterns lead to. For this we will
use the following \XML\ data:
\startbuffer[xml]
c1d1
c2d2
d3
d4
d5
\stopbuffer
\typebuffer[xml]
\xmlloadbuffer{xml}{xml}
\startluacode
function document.ShowResultOfPattern(root,pattern)
local ok = false
for r,d,k in xml.elements(lxml.id(root),pattern) do
tex.print(xml.tostring(d[k]))
tex.sprint(tex.ctxcatcodes,"\\par")
ok = true
end
if not ok then
tex.sprint("no match")
tex.sprint(tex.ctxcatcodes,"\\par")
end
end
\stopluacode
Here come the examples:
\definehead[example][subsubject]
\setuphead[example][style=\tt,before=\blank,after=\nowhitespace]
\def\ShowResultOfPattern#1#2%
{\example{#2}
\startpacked \tttf
\ctxlua{document.ShowResultOfPattern("#1","#2")}
\stoppacked}
\ShowResultOfPattern{xml}{a/b/c}
\ShowResultOfPattern{xml}{/a/b/c}
\ShowResultOfPattern{xml}{b/c}
\ShowResultOfPattern{xml}{c}
\ShowResultOfPattern{xml}{a/*/c}
\ShowResultOfPattern{xml}{a/**/c}
\ShowResultOfPattern{xml}{a//c}
\ShowResultOfPattern{xml}{a/*/*/c}
\ShowResultOfPattern{xml}{*/c}
\ShowResultOfPattern{xml}{**/c}
\ShowResultOfPattern{xml}{a/../*/c}
\ShowResultOfPattern{xml}{a/../c}
\ShowResultOfPattern{xml}{c[@n='x']}
\ShowResultOfPattern{xml}{c[@n]}
\ShowResultOfPattern{xml}{c[@n='y']}
\ShowResultOfPattern{xml}{c[1]}
\ShowResultOfPattern{xml}{b/c[1]}
\ShowResultOfPattern{xml}{a/c[1]}
\ShowResultOfPattern{xml}{a/c[-1]}
\ShowResultOfPattern{xml}{c[1]}
\ShowResultOfPattern{xml}{c[-1]}
\ShowResultOfPattern{xml}{pi::}
\ShowResultOfPattern{xml}{pi::what}
\stopcomponent