GCC/UML/XMI extraction (was gcc front end )

[James Michael DuPont <mdupont777 at yahoo dot com>]

Praveen,

Since no-one has answered your mail yet, I will
attempt to. Note that this is just my personal opinion
on this matter and in no way does it speak for rest of
the gcc community. 

This being my favorite topic, I will explain the
background on this topic and use my answer to present
some of ideas on this topic to you and to the gcc
mailing list. 

>>I'm currently trying to make a tool that'll generate
>>UML specs from c++ code.
I understand your motivation for this request and I
have been thinking about similar things myself.
I too have been sold on the idea of UML and having
nice diagrams of my code. 
"How come we cannot extract this data from the GCC?"
was my the motivating question over 8 years ago.

You would like to extract the class structures of a
given program into UML, maybe into a XMI like data
format?  see : http://xml.coverpages.org/xmi.html
Like many of the UML tools listed here?:
http://directory.google.com/Top/Computers/Programming/Methodologies/Object-Oriented/UML/Tools/?tc=1

That type of extraction or at least the visualization
is one of the one of the goals of the introspector
project that I started. 
The problem is that my project is that XML/SQL or
*any* intermediate representation of gcc tree
structures is a file is not secure enough to prevent
abuse of the gcc.
Therefore I have started a re-design to cover many new
aspects.
http://introspector.sourceforge.net. 
Currently, I am busy re-designing the project to
eliminating all xml and sql external data to remove
the possibility of third-party abuse of the data and
circumvention of the GPL. 

In my humble opinion, In the long term, an UML like
GUI and class browser will be possible to be linked
directly into the gcc and the various aspects of
programs can be visualized. 

You should understand that the pure extraction of such
information by the gcc to feed to case tools like
rational rose and together++ via XMI might not be in
the best interests of the GCC/GNU/FSF community. 

The extraction of compiler data about data structures
and feeding them to non-free software like
GraphVis/Rational Rose/ TogetherC++ and others
does not benefit the free software, but more takes the
wind out of the sails of an effort to create free
replacements for such tools. 

One of the difficulties that we are presented with
creating case tools with the GCC has is the issue of
the usage of the "internal data str." the tree
structures. 

>>I am searching for a library version of GNU's C++
>>front end , and documentation on its interface
>>(through which I can query its internal data str.)
That library is the core gcc compiler, the tree
structures that are connected with the c/c++ language
front ends.

Now in order for you to create such a tool you will
have to create a modified version of the gcc, one that
produces the UML, maybe in XMI format. 

This type of interface would also allow for all types
of information to be extracted from the gcc.

One of the issues involved here is the licensing of
such a tool, it would have to be a GPL licensed tool.

The other issue is that of the extracted data,
are you interested in extracting just static data
structures, or also dynamic UML structures, so that
you would also like to see the run-time behavior of
the program at hand? 

Then you will need to extract also the function
bodies, and from that UML representation, you could
create a brand-new re-factored program.

That is of course a very tempting idea and the basis
of many of the tools like together++ and rational
rose.

There is also a tool the provides such a function for
java written in java called argo-uml, that includes a
Java parser.

There is also a tool called DIA that has support for
UML diagrams. 

In the end both tools could be used for a display
mechanism, argo-uml is much further along than Dia,
but DIA is written in C/C++ and would be easier to use
as an add on to the gcc.

The other question of graph layout can be tackled by a
tool called VCG. That has some good graph layout
algorithms, but is GUI is way out of date.

Those two tools (DIA and VCG), coupled together with
the gcc, statically linked into one program, could
serve as the basis for a C++/UML based code browser
using the gcc. 

If you are only interested in only the static
structure of c++ functions and classes you might be
able to use the xml output from the gcc_xml.
http://www.gccxml.org

>>Could you please guide me to the correct library and
>>where I could download it ?
The gcc parsers and code generators are not
distributed as libraries for linking into. If you want
to create a gcc module, you should create a statically
linked program that uses the data needed.

>>I explored the links on the gcc pages but could not
>>locate these.
It is one of the lesser documented and published areas
of the gcc. The new chapter 18 of the gcc manual
contains a good overview of tree structure.
http://gcc.gnu.org/onlinedocs/gccint/Tree-overview.html

I would suggest that you define exactly your statement
of the how you would like to represent and display the
UML. 
Also you should be willing to put all of your changes
and programs under the GPL and not try and export data
to non-gpled programs. 
Doing that will give you the best basis for such an
endeavor.
It will be a difficult endeavor and a rocky, uphill
battle, but I think it will be worth it. 

Mike 

=====
James Michael DuPont

__________________________________________________
Do You Yahoo!?
Yahoo! Shopping - Mother's Day is May 12th!
http://shopping.yahoo.com