[Pytorch] Add an adapter for c10::string_view

[HGuillemet]

See #1390

I didn't use the owner field of the adapter. I don't think there is a case where we want to free something when the adapter is deleted. Is there ? A string_view is supposed to be a not-owning view of some string.

@sbrunk, could you give that a try ?

[HGuillemet]

I mapped string_view with:

        new Info("c10::basic_string_view<char>", "c10::string_view").annotations("@StringView").valueTypes("String", "BytePointer))

While for std::string it's, by default:

        new Info("std::string").annotations("@StdString").valueTypes("BytePointer", "String").pointerTypes("BytePointer"))

The difference in order of valueTypes results in functions returning string_view return String in Java and functions returning std::string return BytePointer in Java. We'd better unify this.

Do you remember why it's this way for std::string and shall we do the same for string_view ?
Maybe this is to allow application of different encoding ?

[saudet]

BytePointer is more general, yes.

[sbrunk]

@sbrunk, could you give that a try ?

Did a quick test with the original einsum example after building locally. Looking good now :)

println(torch.einsum("ii", torch.randn(Seq(4, 4)))) // -2.5606

[saudet]

@sbrunk Please let us know if this sufficient for all your use cases! Thanks

[sbrunk]

I have to admit I haven't done much testing with other functions using c10::string_view arguments beyond einsum.

This is partly because it's much easier to test with the prebuilt CI snapshots. For instance, right now I'm traveling without access to my Linux build machine and I can't get the preset to build locally on macOS.

So if you merge this now, I can't promise that I won't find any issue once I've improved test coverage. I don't want to cause you any additional overhead though, so if you'd rather like to wait until I've done more comprehensive testing, I'll do it as soon as I'm back/able to build locally again.

[saudet]

Ok, in any case, the build fails on Windows, so that needs to be fixed before we can merge this:

2023-08-17T15:12:19.7007842Z [ERROR] Failed to execute goal org.apache.maven.plugins:maven-compiler-plugin:3.8.1:compile (default-compile) on project pytorch: Compilation failure: Compilation failure: 
2023-08-17T15:12:19.7008718Z [ERROR] /D:/a/javacpp-presets/javacpp-presets/pytorch/src/gen/java/org/bytedeco/pytorch/global/torch.java:[25536,119] cannot find symbol
2023-08-17T15:12:19.7009300Z [ERROR]   symbol:   class StringViewOptional
2023-08-17T15:12:19.7015054Z [ERROR]   location: class org.bytedeco.pytorch.global.torch

[HGuillemet]

The checks should pass now. It passed on linux because the parser is run, and not on windows. Is that on purpose ?