universal script generation for assembly, launcher, and release that run on linux and windows

[lhns]

As a follow-up to com-lihaoyi/Ammonite#781 this pr generates scripts that can run on linux shell and windows cmd.
It merges the shell and batch scripts introduced in #243.
Feedback is always welcome!

[robby-phd]

@LolHens Thanks for following up on this. Not sure if you are done, but I gave it a spin.

I (purposely) tried running the generated mill.bat on macOS, and it gives me:

Error: Could not find or load main class mill.Main
Caused by: java.lang.ClassNotFoundException: mill.Main

I also tried running the generated mill on cmd, and it gives me:

'#!' is not recognized as an internal or external command,
operable program or batch file.
Error: Could not find or load main class mill.Main
Caused by: java.lang.ClassNotFoundException: mill.Main

So it seems the approach that you used for com-lihaoyi/Ammonite#781 is more robust.

Since the script is supposed to be universal for both sh and cmd, can you also remove releaseAll and releaseBatch and change the call to upload in uploadToGithub to use release instead (i.e., similar to before #243)? That way, only mill gets generated (without mill.bat).

Also, can you make shebang defaults to false? We can change the curl/chmod instructions to add the shebang line.

[lhns]

The error you are running into indeed seems very strange. I ran the version I just pushed on my windows system and on a redhat linux machine and it worked fine.
I had to change the approach because before I was prepending a "::" to every line of the shell script so that the cmd skips them however this does not work at all for switch case statements. The new approach tells the cmd to just jump over the shell script and should work for any script.
I removed the releaseAll and releaseBatch and now uploadToGithub only generates "mill".
Also I set the shebang line to be off by default.
Now I am just waiting to see what the tests will say :)

[lhns]

I don't really know why the build failed.

[robby-phd]

Can you try running ci/test-mill-release.sh on your redhat machine using mill 0.1.7-8-b913c6? I got some errors (though different than the ones in travis, which looks spurious).

[lhns]

Unfortunately I don't have access to that machine right now so I ran the tests on a local debian installation but I couldn't get all of the tests to succeed there.
I made a new commit but changed nothing just to restart the build and it seems to have worked where it previously failed at publishVersion in build.sc:360.
I think there might be some other bug involved that is not related to this pr.
Could you try again to execute the generated scripts?

[robby-phd]

I just tried it again locally on macOS and linux, and tests passed (didn't clean out ~/.mill before). The changes look good to me, thanks.

[lhns]

Very nice, thanks! Should I make a pr to change the generated script in ammonite according to this so that the code is equivalent?

[robby-phd]

That would be great!

[lihaoyi]

@robby-phd it seems as of this commit we stopped uploading windows bat files to Releases. Any idea what happened?

[robby-phd]

@lihaoyi The published mill is now both a batch and sh script (universal) so we don't need to publish separate mill releases.

Windows users only need to save mill releases as mill.bat. I have documented this in the master changelog (and in #234 for 0.1.8 release).

[lihaoyi]

Oh ok, got it