22 year old undergraduate student of architecture at Indian Institute of Technology, Kharagpur.
GSoc 2009 Proposal Draft
Continuing with the libpc work from last gsoc, the proposal is for providing the following functionalities in mged/libged for the end-user
- creation/modification of constraint objects
- a solve command for solving a set of constraints / all the constraints in the loaded .g file
- options for selection from a range of solutions if possible
for achieving the same, the work falls under the following heads
- An inclusive definition of constraint grammar on the basis of the existing Math grammar which can be easily extened later.
- Integration with libged - framework for calls into librt and libpc for solution, storage and updation.
- Writing a robust modular constraint solver
- Writing implicit constraints for the primitives : Using libpc for the rt_prep checks.
- Final Test: Testing explicit constraints which would involve checking the list of commands to be implemented by the proposal thereby resulting in a complete test of the framework.
The status of libpc project as of now is as follows.
- Abstraction Framework - Objects implemented : Parameters, Variables and Constraints, Constraint Network
- Solution System - Generic solver testing all possible variable assignment combinations, Graph specific solver using boost graph library
- Math Virtual Machine (work in progress) - expected to be complete with around 400 to 500 lines of code change at max.
- Basic constraint I/O in librt - will need changes till constraint grammar is finalized
In the long run, I am sure that the rigor of the solver would be the major factor in terms of usability as well as performance. Even though I have been toying around with some generic constraint solvers as well as geometric ones, I haven't found one which could be easily integrated into brl-cad. The ideal approach would be to make the solver an independent module. This would not only support using multiple solvers by using the same "protocol or format", but also simplify the process of progress in terms of solver optimizations.
I believe a good project definition for the approximately 480 coding hours of gsoc ( May 23rd to August 10th) is as follows.
1. Solution system (~170 hours)
Undoubtedly the most important aspect of the constraint system. Two approaches can be taken for modularity of the solver from other parts of libpc. One way would be to use a server-client model possibly using inter-process communication or a simple file based transfer of data between the two. Alternatively it could be a part of the libpc library itself requiring a particular format of the constraint satisfaction problem. This could be passed on to the solver as string data or as a data structure. Of the various approaches towards solving geometrical constraints, I am slightly in favour of graph-constructive methods over the others (see #1) But eventually I think something like a Locus Intersection method (#4) which is a hybrid between the computational and search approaches would be ideal. One of the reasons I feel that the solver should be completely independent of the constraint satisfaction problem definition or representation is the fact that it is a field rapidly undergoing changes in terms of new approaches.
2. Constraint grammar (~100 hours)
This would be dependent on the underlying Math grammar system. Testing out the grammar system would include correct parsing of expressions implying the constraints into the constraint objects. For instance ( " tangential(ell1, sph1)" being parsed into the correct constraint objects containing evaluating functions in terms of the variables)
3. Test Use-cases for Implicit Constraints (~80 hours)
This involves using the libpc framework for doing the rt_prep checks for each of the geometry primitives. This work can be pursued independent of the work on the solution system , using the existing generic solver. Also in terms of implementation this would be a good start to finalize the common type of constraints grammar wise ( parallelism, perpendicularity etc.)
4. Test Use-cases for Explicit Constraints (~130 hours)
With the wrapping up of Math Virtual Machine and grammar, this work is effectively the test of the entire framework and would involve purely construction of test cases.
In addition the following also need to be done.
- Some tinkering is also needed to make the existing code more modular.
- There is also a "small" memory leak in the solver which needs to be taken care of.
In the meantime, I will try to wrap up the documentation which needs a major overhaul and more importantly finish off with the MathVM completion which I have started in March.
May 23 - July 6 (~ 6 weeks)
- Do further groundwork on Solver
- Finalising Constraint grammar
- Work on Implicit Constraints
- Initial solver work
July 6 - August 10 (~ 5 weeks)
- Complete Solver work
- Test Explicit Constraints
There are many aspects of libpc which I have not taken into consideration in this proposal , for instance the integration with libged so that the whole system becomes accessible to the end user. But I believe the above mentioned aspects need to be taken care of first for any progress. Considering my last year experience, I still think this is probably a little more than what I should really aim for.
 A geometric Constraint Solver, W Bouma, I Fudos, C Hoffmann, J Cai, R Paige - Computer Aided Design, 1995
 A Modular Geometric Constraint Solver for User Interface Applications, Hiroshi Hosobe , ACM symposium on User interface software and technology, 2001
 Solving geometric constraint systems, GA Kramer, BB Qh - 1992]
 Solving spatial basic geometric constraint configurations with locus intersection, XS Gao, CM Hoffmann, WQ Yang - Computer-Aided Design, 2004 - Elsevier doi:10.1016/S0010-4485(03)00056-3