Difference between revisions of "User:Tandoorichick/GSoC2016/Logs"

From BRL-CAD
(Development Logs)
(Development Logs)
 
(8 intermediate revisions by the same user not shown)
Line 119: Line 119:
 
21/7/16 - Came up with results.
 
21/7/16 - Came up with results.
  
22/7/16 - Fell ill again
+
22/7/16 - 25/7/16 - Fell ill again
  
 
26/7/16 - Made changes according to Daniel's suggestions.
 
26/7/16 - Made changes according to Daniel's suggestions.
 +
 +
27/7/16 - 11/8/16 - Had fallen ill and couldn't get to work.
 +
 +
12/8/16 - Outlining skewed lines support for 3-D lines (while calculating feature).
 +
 +
13/8/16, 14/8/16 - Debugging the instance of the heal command being called on a mesh on which the heal command had already been called.
 +
 +
15/8/16 - Debugging the coplanar condition addition while finding the closest edge.
 +
 +
16/8/16 - Debugging errors in making the DCEL, with a torus.
 +
 +
17/8/16 - Testing with meshes to figure out the point of breaking. Wrote blogs for weeks missed.
 +
 +
18/8/16 - Started with taking tolerance as user input. Errors popped up. Could not identify what was causing them even after falling back to old code.
 +
 +
19/8/16 - Build error resolved. Added feature to take in zipper tolerance as user input. Debugged with Daniel's suggestions. Error seems to be with setting the incident face reference - because of the orientation checking. With the spheres the origin was within the whole sphere, so orientation was consistent. With the torus, the origin is not inside the object, so inconsistencies with orientation don't allow some edges to get the right incident face. Hence, the next and previous references get messed up too. Yet to think of a fix.
 +
 +
20/8/16 - Dealing with the mesh not being completely healed because some vertices don't get valid feature pairs, even while having them. Added a special case for when the hole is triangular to reduce compute time.
 +
 +
21/8/16 - Had missed a condition in checkAndSetTwins() routine. Debugged and found it. This prevented the case where, when a mesh is healed with the heal command, and the heal command is called again on the mesh, the algorithm hung. This was because of the presence of singular vertices which was in turn because the mesh was not completely healed. This was rectified.

Latest revision as of 10:55, 22 August 2016

Blog with week-wise work including all ideation and explanation at: http://tandoorichick.github.io/

Development Logs[edit]

23/5/16, 24/5/16 - Confusion regarding the indexed and unindexed format of BRL-CAD and OpenSCAD -> researched other ways to implement the portable module. Confusion regarding the use of DCEL as well.

25/5/16 - Outlined the PolygonalMesh abstract class. Listed out the functions that will be required for the conversion from PolygonalMesh to the DCEL.

26/5/16 - Outlined the doubly-connected edge list class. Declared virtual functions (that will obtain information from the intermediate mesh to convert to the DCEL) in the PolygonalMesh abstract class. Others will be added later on when the need arises.

27/5/16 - Set up the BrlcadMesh class and defined functions. Implemented few geometry utility functions - orientation checking. Went through source to find function function: rt_bot_same_orientation() to check if an edge and a triangle are oriented the same way.

28/5/16 - Vertex record functions definition

29/5/16 - Face record functions definition

30/5/15 - Edge record functions definition

31/5/16 - Setting up OpenSCAD and going through the code. Little bit confusion regarding the Vector3D variable.

1/6/16 - Implemented some geometry util functions - checking if two triangles are oriented the same way, calculating determinant. Face record functions

2/6/16 - Vertex record functions and some of edge record functions

3/6/16 - Edge record functions

4/6/16 - Classes' design changes. Functions common to both native structures were implemented in the PolygonalMesh class itself.

5/6/16 - UI commands support - added a dummy command "heal" and tested with a sample message

6/6/16 - Changes in the conversion module code according to Daniel's suggestions

7/6/16 - Defined the queue element for the zipper gaps algorithm, set up the files and started with outline.

8/6/16 - Wrote functions for initializing priority queue and listed out other utility functions that might be needed

9/6/16 - Function to zipper gaps

10/6/16 - Implemented utility functions - finding orthogonal projection, calculating feature pair, etc.

11/6/16 - Out of station

12/6/16 - Out of station

13/6/16 - Set up the module inside the libanalyze folder in the BRL-CAD source. Functions to update the DCEL structure - like splitting a face, inserting a vertex on an edge

14/6/16 - Created a separate header Geometry.h to keep all geometry related functions like finding if a point can be orthogonally projected to a line, find such a projection, distance between two points, distance between a point and a line, etc. Continued writing functions to update the DCEL structure and inserted calls to them.

15/6/16 - Wrote utility functions for accessing and updating the rt_bot_internal structure.

16/6/16 - Had set the unbounded face id to be the number of faces initially, changed that after facing problems while adding and deleting faces. Updated all relevant functions. Came up with an approach for finding the closest edge that does not cross any existing vertex/edge/face. Solved other errors. Built code successfully.

17/6/16 - Had to write TCL command binding again (because i had checked out revision without saving the work). Analysed the closest edge function logic.

18/6/16 - Personal commitments

19/6/16 - Moved header files around to include directory, but undid. Build errors due to missing headers when i included a header file in the existing code. Asked for help on the IRC.

20/6/16 - Wrote analyse_heal_bot to link to the TCL command. Function to find free edge chains that are not loops was also written. Including C++ headers in the MeshConversion.h file gives errors while including MeshConversion.h in src/libwdb/bot.c for analyse_heal_bot(). Yet to figure out. Went through earlier code to correct any apparent mistakes.

21/6/16 - Modified code for finding closest edge. And modified code to conform to BRL-CAD coding standards. Uploaded modified patch on SourceForge. Started work with stitching gaps since compilation was unsuccessful. Got solution, made changes. Yet to compile successfully.

22/6/16 - Error while linking C++ code to C. Started with stitching in the meanwhile. Compiled module later on. Started debugging.

23/6/16 - Tested code. Conversion module gets DCEL to shape. Testing of zippering gaps going on.

24/6/16 - Changed logic for closest edge function, to handle feature edge and feature vertex properly. Works for most cases. Rest yet to be fixed.

25/6/16 - Figured out way to separate the boundary of a mesh from the defects (using orientation). Till initization of the priority queue works fine (finding free edge chain, finding closest edge for those, calculating error and inserting into the priority queue. Memory corruption error with the zipperGaps() function. I'm trying to resolve it.

26/6/16 - Resolved one memory corruption error. Another popped up with the priority queue. Finding ways to resolve that.

27/6/16 - Resolving memory corruption errors

28/6/16 - Made changes in the zipperGaps() function. Tested them. Errors in logic still exist in some of them. Looking into it.

29/6/16 - Errors with the functions modifying the DCEL. Rectifying them.

30/6/16 - Still modifying the DCEL access functions. Wasn't well for a part of the day.

1/7/16 - Errors with function addEdge rectified. Setting references while deleting edge modified.

2/7/16 - Rectified errors with delete functions. Edge contraction and vertex contraction processes working fine.

3/7/16 - Corrected minor errors.

4/7/16 - Next and previous edge references set right after the contractions. Sorted out memory issues, changed function prototypes accordingly. Parameters with double pointers changed to reflect that they are 3D arrays. Undefined behavior w.r.t even memory issues seen. Trying to resolve. Started coding stitching algorithm.

5/7/16 - Memory corruption issue wrt push operation with priority queue. Looked for solutions. Couldn't find any fix. Coding the stitching algorithm. Made changes from yesterday - that both chains move in opposite directions by virtue of being a part of the same loop.

6/7/16 - Coded the rest of the stitching algorithm. Included a part in the zippering gaps algorithm to accommodate for gaps that are square in shape (in which case none of the vertices on the chain would have any valid feature - hence it needs to be treated separately). Changed some function parameters to const as Daniel had suggested.

7/7/16 - Tried fixing the memory corruption error with the priority queue, but couldn't. Check the stitching algorithm code. Went through the overlaps fixing algorithm for coding.

8/7/16 - Held up with some other work. Thought about changes to the existing zipper module code, and the stitching algo.

9/7/16 - Zippering gaps modified to include boundary edges. bool vector is used to avoid multiple checks to the same chain of edges. Stitching gaps algo modified slightly. Zippering - vertex contraction has not been done right. Debugging.

10/7/16 - Weekend

11/7/16 - Issues with vertex contraction resolved. Facelist starting edge references are out of place. Degenerate cases wrt orientation of triangles causing those issues. Changes to stitching gaps module outline.

12/7/16 - deleteEdge() function changed to cope with the degeneracy wrt orientation. Works fine. Checking the functions that might be causing the memory issues. Added changes to stitching algorithm.

13/7/16 - Memory issues resolved. Found cases where triangles might cross over each other due to the transformations. Thus, need to check if the orientation of the triangle changes. Changes made in the rt_bot_internal struct pointer not being reflected with draw. Need to figure out.

14/7/16 - Changes could be seen. But with distorted triangles. Changes for orientation check was added. Works. Few more cases of erroneous detection of feature edges, looking into them.

15/7/16 - Zippering gaps module works fine. Need to test with other meshes where the tolerance and the size of the triangles is not so comparable.

16/7/16 - Cleared out dead code. Wrote includes properly. Added a bit of stitching code.

17/7/16 - Weekend

18/7/16 - Had to test a few times with a mesh that needs no healing. Made changes for the case when the defect is a square (the normal case does not work, special case was written for a square) - works fine. Created two disconnected hemispheres on Blender to test, and exported the geometry as an .obj file. Testing it.

19/7/16, 20/7/16 - Was not well. Couldn't spend much time. Fixing two hemispheres works okay, but needs more testing.

21/7/16 - Came up with results.

22/7/16 - 25/7/16 - Fell ill again

26/7/16 - Made changes according to Daniel's suggestions.

27/7/16 - 11/8/16 - Had fallen ill and couldn't get to work.

12/8/16 - Outlining skewed lines support for 3-D lines (while calculating feature).

13/8/16, 14/8/16 - Debugging the instance of the heal command being called on a mesh on which the heal command had already been called.

15/8/16 - Debugging the coplanar condition addition while finding the closest edge.

16/8/16 - Debugging errors in making the DCEL, with a torus.

17/8/16 - Testing with meshes to figure out the point of breaking. Wrote blogs for weeks missed.

18/8/16 - Started with taking tolerance as user input. Errors popped up. Could not identify what was causing them even after falling back to old code.

19/8/16 - Build error resolved. Added feature to take in zipper tolerance as user input. Debugged with Daniel's suggestions. Error seems to be with setting the incident face reference - because of the orientation checking. With the spheres the origin was within the whole sphere, so orientation was consistent. With the torus, the origin is not inside the object, so inconsistencies with orientation don't allow some edges to get the right incident face. Hence, the next and previous references get messed up too. Yet to think of a fix.

20/8/16 - Dealing with the mesh not being completely healed because some vertices don't get valid feature pairs, even while having them. Added a special case for when the hole is triangular to reduce compute time.

21/8/16 - Had missed a condition in checkAndSetTwins() routine. Debugged and found it. This prevented the case where, when a mesh is healed with the heal command, and the heal command is called again on the mesh, the algorithm hung. This was because of the presence of singular vertices which was in turn because the mesh was not completely healed. This was rectified.