Experiment git-submodule for data-set

[sloriot]

• Add all data set in a dedicated package, this should not break the release creation or testing
• Update cmake scripts so that in branch-build testing a package is still working
• Make the data set package a git-submodule
• Update release creation script to accomodate git-submodule
• Document how to use git-submodule for the data-set
• See how easy it would be to add another submodule with other data-set that would also bring new tests
• Split the data set package for tests and examples+demos (only one out of the two ends up in a release) ?

[sgiraudot]

Do we want to stop separating data sets from one package to another? (I'm not sure if this is how you see things, would it be something like Data_sets/data/one_data_set.off or Data_sets/data/Polygon_mesh_processing/one_data_set.off?)

I think it would be good to dump the package separation, because some packages actually duplicate data (like point set processing and point set shape detection). Maybe we could still separate the data files (in order to avoid having one directory with hundreds of files) by their types? Point sets, polygon meshes, etc.?

[sloriot]

For example PMP/test/PMP/data/abc.off would be moved to Test_data/test/PMP/data/abc.off, and PMP/examples/PMP/data/abc.off would be moved to Release_data/examples/PMP/data/abc.off.

[efifogel]

First, I think that one repository called cgal_data should suffice. (This may not conflict with previous posts.) Secondly, I think we should allow sharing data (i) between examples, tests, and demos, and (ii) between different packages. How about the following: Each set is associated to a pair of a specific package and a specific type of app. (either examples, tests, or demos), so for example: cgal_data/polygon_2 cgal_data/polygon_3 cgal_data/polyhedron_3 cgal_data/point_2 cgal_data/point_3 cgal_data/segment_2 cgal_data/segment_3 cgal_data/surface_3 cgal_data/circle_2 cgal_data/sphere_3 The problem is that a certain set can become huge, while only a small portion of it is used by specific package/app-type pairs. In this case, we can split it either \ cgal_data/polygon_xxx_2 or cgal_data/polygon_2/xxx The rest can be handled with scripts...

…

____ _ ____ _ /_____/_) o /__________ __ // (____ ( ( ( (_/ (_/-(-'_(/ _/

On Fri, Jan 6, 2017 at 12:44 PM, Sebastien Loriot ***@***.***> wrote: For example PMP/test/PMP/data/abc.off would be moved to Test_data/test/PMP/data/abc.off, and PMP/example/PMP/data/abc.off would be moved to Release_data/test/PMP/data/abc.off. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1803 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGoWuL7k0C9CHELcpPyzV5AzG3KkZHE3ks5rPhr2gaJpZM4LclCq> .

[maxGimeno]

Why do we need to keep the package separation ? I agree with Simon, it would be easier to use and it would avoid data duplication if all the files were in the root cgal_data directory.

[sloriot]

That was a first step as it does no require to heavily modify scripts (that is to focus on the technical git-submodule part).