Programming language + XML + XSL = compiler ?

[rnbrady]

Hi folks

New around here. This post may be a duplicate, I think I bungled the
previous one.

I've recently started out with XML and XSL and I'm very impressed with
both. My use is non-web in nature. In fact, I'm writing a compiler of
sorts, from high-level spec in XML, with eventual generation of VHDL.

I've made the following observations:

1. Given pure source code (Eg:Java,C++), one could generate an abstract
syntax tree in XML.
2. XML is also apposite for symbol tables and intermediate froms.
3. XSLT can do powerful transformations of said internediate forms.
4. XSL can generate plain text.

I'm wondering wondering what it would take to implement a compiler for
a given language in XML/XSL. More specifically, are there any XSL or
XML based systems that assist in scanning pure source into XML so that
the transformations can begin? Grammar could be provided in marked-up
EBNF. This is probably over-optimistic could the process would need
knowledge of LL/LALR etc.

The alternative approach is to use your standard scanner/parser
generator with it's own (non-interchangable) EBNF and have the action
code generate tagged tokens, giving you an AST in XML.

My questions:
1. Are there any decent XML shema's for EBNF?
2. Are there any tools / XSL extensions to do the above?
3. What are your thoughts on the whole idea?

Best regards,
Richard Brady

 

[Joe Kesselman]

I'm wondering wondering what it would take to implement a compiler for
a given language in XML/XSL.

Remember, XSL is a pure functional langauge. That may make some of the
manipulations a serious compiler would want to perform difficult.

But, yes, a basic old-fashioned just-replace-with-equivalent-code
transformation ought to be possible. This is not the set of tools I
would choose to do it in -- custom data structures are likely to be much
more efficient than trying to model everything in XML -- but if you're
looking for a pedagogical exercise, it could be a fun thing to try.

Were I you, I'd start with a language that has simpler syntax and
semantics than Java or C++, and use that to establish your framework and
understand the issues (including performance and memory space). Then, if
you're still feeling ambitious, go for it.

(Personally, I've stopped writing compilers for the sake of writing
compilers. I'd rather wait until I'm reasonably sure there's something
mine will do better than any existing alternative.)



More specifically, are there any XSL or

 

[Andy Dingley]

3. XSLT can do powerful transformations of said internediate forms.

IMHE, XSLT can't do "powerful", for interesting values of powerful. It
does "Bulky but simple" very well indeed and is excellent for text
processing with blocks of text. If you try and do anything involving
semantics though, it starts to fall flat.

 

[=?ISO-8859-1?Q?J=FCrgen_Kahrs?=]

I'm wondering wondering what it would take to implement a compiler for
a given language in XML/XSL. More specifically, are there any XSL or

This has been suggested 5 years ago in

http://www.idealliance.org/papers/xml2001/papers/html/03-05-04.html

and also in comp.compilers late last year.
Google for "xml as intermediate representation"

3. What are your thoughts on the whole idea?

Teaching compiler construction could be an area
where this idea might be useful. But in real life,
I would _not_ like to write my source code in XML.

 

[Joe Kesselman]

Then there's always the inverse: a programming language whose syntax is
based on XML, so it can be styled by XSLT (or generated from other XML
sources by XSLT). IBM's Bean Markup Language and BML Scripting Pages,
which I was involved in some years ago, were one such example, though in
that case the language was designed for configuration and assembly of
existing Java components (and BSF scripts, also something that I helped
develop) rather than being intended as a completely stand-alone
programming environment.

 

[Andy Dingley]

Then there's always the inverse: a programming language whose syntax is
based on XML, so it can be styled by XSLT (or generated from other XML
sources by XSLT).

Ant

SQL DDL, as managed by the Apache Torque project

 

[Joe Kesselman]

Hm. I don't usually think of Ant as a programming language, and I don't
think I've heard of anyone styling it, but that's an interesting
observation -- stylesheets to adapt Ant builds to different
environments, perhaps...

 

[rnbrady]

Hi folks

Thanks for the input. I'm not actually thinking of doing this as a
project, just curious about whether any investigation has been done on
the topic. I currently use this technique to transform digital signal
processing (DSP) algorithms specified in XML into VHDL implementations
of those algorithms.

Perhaps I'm off the mark with suggesting it for C++ or Java, but it's a
particularly interesting concept when considering higher-level
specifications such as in model driven engineering (like UML in MDA). A
large emphasis becomes interoperability/exchangability/portability of
models, but at the core of it all is transformations between models.

It's interesting because we all have different backgrounds and
therefore unique perspectives on an concept like this.

Thanks again,
Richard

 

[Juergen Kahrs]

the topic. I currently use this technique to transform digital signal
processing (DSP) algorithms specified in XML into VHDL implementations
of those algorithms.

Where did you get the DSP algorithms in XML ?
Is there any standard for this ?

 

[Andy Dingley]

Hm. I don't usually think of Ant as a programming language,

In its limited little way, it's a declarative programming language (as
was Prolog). You can use it as a convenient dependency resolver.

I don't think I've heard of anyone styling it,

I have an XSLT stylesheet that auto-documents Ant external targets.
Handy when I'm shipping deployment scripts to users.

I also auto-generate Ant from a vile combination of our bug tracker,
change control, and Torque repository. This generates database patch
scripts for upgrading customer databases from arbitrary start points to
new product versions.

 

[Joe Kesselman]

Another interesting reference on this topic: The folks working on a test
suite for the W3C's DOM APIs were developing a limited XML-based
"programming language" for binding/language-independent description of
the tests, together with stylesheets which would render those into
executable form. Again, that's a fairly primitive "compiler", but it's
yet another instance of XML-based programming languages.

 

[rnbrady]

Juergen

Where did you get the DSP algorithms in XML ?

That's a good question. I'm writing them myself and learning as I go
along.

Is there any standard for this ?

Not that I know of, but I wouldn't be surprised.

So, now that I have these XML descriptions being converted into VHDL,
the next step is to layer something more interesting on top. No DSP
designer wants to type their spec in XML ... yuk!

There are several options: text (something more readable than XML),
visual, or both. My application is Software Defined Radio so I'm
looking at Waveform Description Language (WDL) which has a textual form
with equivalent representations in visual and XML forms. Another option
would be UML2.

For WDL I have a Lex/Yacc grammar, so I can just create a little
compiler to mark-up the code. I thought there'd be plenty other folk
with the same issue and some standard solution might have been
developed. I'm a big fan of standard solutions because of the re-use
which they encourage.

Richard

 

[rnbrady]

Juergen

PS: Your links (both the article and the comp.compiler discussion) were
very useful. Thank you.

 

[=?ISO-8859-1?Q?J=FCrgen_Kahrs?=]

So, now that I have these XML descriptions being converted into VHDL,
the next step is to layer something more interesting on top. No DSP
designer wants to type their spec in XML ... yuk!

Two years ago, someone posted here about a VHDL
application. Use Google Newsgroups to search for
"Use variables to get unique nodes". He posted
this stuff:

<?xml version="1.0" encoding="UTF-8"?>
<Root>
<E>
<!-- If we have the key match with pin only
the xslt does not work. If the maych is at least TEST/Pin
it is OK
-->
<Pin PinName="A" PinClass="T"/>
<!--<Pin PinName="AX" PinClass="T"/> -->
</E>
<x>
<!-- in case the TEST node below is not selected by Index
AND PinName="A" and CellType="M" matches
the result is wrong
-->
<TEST Index="42" CellType="M">
<Pin PinName="z" PinClass="R"/>
<!--<Pin PinName="AX" PinClass="T"/>-->
<Pin PinName="A" PinClass="T"/>
</TEST>
<TEST Index="43" CellType="M">
<Pin PinName="A" PinClass="T"/>
<Pin PinName="D" PinClass="R"/>
<Pin PinName="C" PinClass="R"/>
<Pin PinName="X" PinClass="R"/>
</TEST>
<TEST Index="44" CellType="M">
<Pin PinName="D" PinClass="R"/>
<Pin PinName="C" PinClass="R"/>
<Pin PinName="X" PinClass="R"/>
<Pin PinName="A" PinClass="T"/>
</TEST>
<TEST Index="45" CellType="M">
<Pin PinName="z" PinClass="R"/>
</TEST>
</x>
</Root>

There are several options: text (something more readable than XML),
visual, or both. My application is Software Defined Radio so I'm
looking at Waveform Description Language (WDL) which has a textual form
with equivalent representations in visual and XML forms. Another option
would be UML2.

This is rather abstract. You should post an example.

 

[=?ISO-8859-1?Q?J=FCrgen_Kahrs?=]

PS: Your links (both the article and the comp.compiler discussion) were
very useful. Thank you.

This paper about lcc may also be interesting for you:

http://www.cs.princeton.edu/software/lcc/asdl.html

This is not strictly involving XML, but they also
use a tree-representation.

 

[rnbrady]

Thanks!

 

[rnbrady]

I recently came across the str:tokenize() function at the EXSLT
project. It takes a string and converts it into several tokens.

QUOTE:

The string given by the first argument is split at any occurrence of
any of these characters. For example:

str:tokenize('2001-06-03T11:40:23', '-T:')


Gives the node set consisting of:

<token>2001</token>
<token>06</token>
<token>03</token>
<token>11</token>
<token>40</token>
<token>23</token>

UNQUOTE.

This is close to what I had in mind. If you could extend the idea so
that it takes an EBNF in XML form instead of a list of delimiters,
wouldn't you have part of a scanner and parser?

 

[=?ISO-8859-1?Q?J=FCrgen_Kahrs?=]

This is close to what I had in mind. If you could extend the idea so
that it takes an EBNF in XML form instead of a list of delimiters,
wouldn't you have part of a scanner and parser?

Yes, but what is it good for ?
Do you want to write a universal Parser which
takes an EBNF and parses according to the EBNF ?
Such universal parsers have been suggested and
implemented much earlier. For example, Niklaus
Wirth presented such a parser in 1976:

http://www.inf.ethz.ch/personal/wirth/books/Compilerbau0/
http://www.oberon.ethz.ch/WirthPubl/CBEAll.pdf

 

[rnbrady]

Do you want to write a universal Parser which

takes an EBNF and parses according to the EBNF ?

It doesn't need to be written from scratch, just integrated as an
extension to XSL, so that XSL can take structured text (in a non-XML
form) as input and parse it, and generate xml/html/text as output.

Yes, but what is it good for ?

1. Compiler-compiler not needed as separate step in the whole process.
2. Definition of a standard syntax for EBNF using a DTD or XSD.
3. Entire compiler could be written in XSL.
4. An interesting platform for academic exploration of compiler design.

 

[Philippe Poulard]

In its limited little way, it's a declarative programming language (as
was Prolog). You can use it as a convenient dependency resolver.




I have an XSLT stylesheet that auto-documents Ant external targets.
Handy when I'm shipping deployment scripts to users.

I also auto-generate Ant from a vile combination of our bug tracker,
change control, and Torque repository. This generates database patch
scripts for upgrading customer databases from arbitrary start points to
new product versions.

hi,

I have designed and implemented a general-purpose XML system that I
called "Active Tags" which allows to make programs with XML tags ; it
allows to mix declarative tags with imperative instructions, uses XPath
to access object's properties, and much more...

it is not "yet another markup language" but a framework that allows
several markup languages to run together

the implementation (Java) :
http://reflex.gforge.inria.fr
the specs :
http://disc.inria.fr/perso/philippe.poulard/xml/active-tags/active-tags/active-tags.html
http://disc.inria.fr/perso/philippe.poulard/xml/active-tags/

--
Cordialement,

///
(. .)
--------ooO--(_)--Ooo--------
| Philippe Poulard |