In general seems like it can be nice functionality to have.

Some concerns.

• the chance of looping forever is risky. It would be good if blender has some way to press Escape to cancel long running python scripts.
• this looks like it intends to operate on the entire mesh, so probably it should un-hide before beginning.

To better handle cases of really bad mesh data which possibly can't be fixed, I think the loop that checks while non_manifold_vertices > 0: It should really check if any changes where made as well.

So rather then looping while there is some manifold geometry, It should loop and keep fixing, as long as it can keep making fixes, Once no changes can be made - it can break out of the loop.

Other comments,

• Recently add options for select manifold, probably its good to check on these: https://developer.blender.org/rB00cb9e5f1cfab142b0972f7a35c3247b00d56423#7dfccffc
• bpy.ops.mesh.remove_doubles(use_unselected=True) why is use_unselected set? This is meant for merging unselected into selected, which doesnt make sense in this case?
• bpy.ops.mesh.fill() takes a sides options, defaults to 4, probably good to set this. (and perhaps you want to set it higher? - not sure)
• bpy.ops.mesh.dissolve_degenerate and bpy.ops.mesh.select_all take a threshold argument. Probably good if this operator defines a threshold and passes it to all operators it calls which also take a threshold.

Thanks for reviewing!
I did a pretty big refactor, to try to address the comments you brought up.

I now terminate the loop if it exceeds a max iteration as well as terminate it if nothing was fixed in a particular iteration.

The operator now takes some arguments for threshold and default sides to fill.

I saw the options were added for selecting manifolds, but I didn't see how to specify the option in the API docs.
[[ http://www.blender.org/documentation/blender_python_api_2_71_release/bpy.ops.mesh.html?highlight=manifold#bpy.ops.mesh.select_non_manifold

Hi, thanks for the update, I had a look and made some changes, interested to know if they work for you.

Summery of changes

• was switching from editmode to object mode just to count selected vertices, now stay in editmode and simply count vertices.
• use different method of checking when to break the cleanup loop:

• removed some unused functions.
• add self.delete_loose() once at the start (any reason you removed?)
• postpone make_normals_consistently_outwards until after all cleaning is done.

It looks great!
Thanks!
I definitely learned a few things from reading your code. :)

I think I accidentally deleted delete_loose().

I was worried about postponing make_normals_consistently_outwards, since I wasn't sure how edges in between faces with opposite normals would count, but it looks fine.

I tested it on Rakdos. It's the model that more naive repairs out there fail on. I can't attach the STL (or even a zipped STL) because it exceeds the size limit.

Do you think using unique (totvert, totedge, totface) will get false positives? Like the suppose the total number of vertices stayed the same, but the non manifold vertices themselves have changed (through some combination of fill() and delete). I'm not sure if this is likely or even possible.

Regarding, make_normals_consistently_outwards - your right actually, I'll need to modify this to disable use_non_contiguous option, this way edges will be ignored between flipped faces. Probably theres no use in having use_multi_face option enabled either (3+ edges to a face).

Would you be able to upload the test file and link the URL here? - so I can check my edits work ok.

Re: (totvert, totedge, totface) giving false positives, think the likelyhood is so low as to not being worth worrying about (in fact impossible at the moment).

• Vertices are always removed (never added).
• Faces are always added (except when verts are removed that collapse faces).

... so I cant see chance that both stay same when an edit is made.

I've uploaded the files to GitHub. It's not a huge sample size, but I think it's somewhat representative of the cases.

https://github.com/caretdashcaret/MeshRepairTestModels

@Jenny Cheng (caretdashcaret) - to be clear, does the current version of this patch work as you'd expect?

Yup! The patch worked great!

I think you've also addressed my concerns.

Committed rBAa88a2e646018b3f9b8d1f818f0a08370b2a0dd67
With some edits

• Control whats selected by select_non_manifold_verts.
• Report a message showing verts added/removed on completion.
• use staticmethods/classmethods (so functions only use whats passed to them)