@gottesmm could you take this up? Can you add Benchmark_O --version parameter that would report the same info as swift --version? I have looked at cmake a bit, but I’m not clear on how to make it work. It looks like you have more experience in the area...

@palimondo I am working on other things right now. Let me explain to you how to do this if it interests you. I think this would be another one of those really nice solid contributions to get under your belt.

I do not think that one can programmatically AFAIKT get this information (but I may be wrong). That suggests to me we should go down the cmake configure route. To see how to do that, please first read this section of the cmake guide:

https://cmake.org/cmake/help/v3.4/command/configure_file.html

and look at these examples from the benchmark suite itself where we have accomplished such things.

https://github.com/apple/swift/blob/master/benchmark/scripts/CMakeLists.txt

There are a few challenges here that we need to consider:

1. How would this work in swiftpm? At least this shouldn't break swiftpm.

2. Given that we support out of tree benchmarking, I would rather the benchmark suite be consistent and just use the version output from the swiftc that it is using to compile the sources. That can be done via runcmd: https://github.com/apple/swift/blob/master/benchmark/cmake/modules/SwiftBenchmarkUtils.cmake#L53

Your thoughts?

So far I got a Version.swift.in that produces Version.swift using the configure_file. Any tip on how to make this work with SwiftPM?

This is my first look at SwiftPM, so… looking at Package.swift, it looks like I can run arbitrary code there? Would it be OK to shell out there with swiftc --version and generate Version.swift myself? Basically reimplementing runcmd and configure_file from cmake side?

There are a bunch of ways we can do this. I am really not sure what is the "right way". Lets ask the SwiftPM people if there is a better way to do this: @aciidb0mb3r.

@palimondo I talked with some people and some further thoughts:

1. I think for the swiftpm case it is ok to just output nothing or at least a message that says N/A. SwiftPM will eventually have a feature to do this but they do not today.

2. The information that you really want is swift version and benchmark version. That to me says that we want swiftc --version as a human readable string and to get the actual git hash of the benchmark suite.

Let me add a bit more information here. When I spoke with the swiftpm people they were pretty against shelling out. That is why I am suggesting that we just do something with swiftpm that doesn't print out the version. Since it is only for editing at this point, it is fine to do.

Umm… I think this still leaves me with a problem: To make the benchmark compile, I have to have Version.swift file present for the build. I'd be generating that with cmake as you taught me from examples above using the Version.swift.in, with only latter being in the version control. But SwiftPM requires me to have Version.swift under version control too… so I would always be leaving the Verision.swift in modified state as far as git is concerned. Wouldn't that we less than ideal? Am I missing something?

I would have a separate Version.swift for swiftpm.

Let me elaborate slightly. With cmake we have much more control vs swiftpm. So I imagine it would work like this:

1. In the cmake case you configure Version.swift.in into the build directory and teach cmake how to find it. CMake won't know anything about SwiftPMVersion.swift.

2. SwiftPM will just pick up SwiftPMVersion.swift by default if you put it in the utils directory.

OK, I think I got this (in principle). I'm stuck on building with SwiftPM. It always crashes on me.
https://gist.github.com/palimondo/1af3c794ef2efe17bbb15e71127a61aa

@gottesmm can you get the benchmarks to build with SwiftPM?