Scala runner doesn't do executable jars

[som-snytt]

Compiler version

3.0.0

Minimized code

➜  snips sdk use scala 2.13.6

Using scala version 2.13.6 in this shell.
➜  snips scalac -d hi.jar hi.scala
➜  snips scala hi.jar
hello, world
➜  snips sdk use scala 3.0.0

Using scala version 3.0.0 in this shell.
➜  snips scalac -d hi.jar hi.scala
➜  snips scala hi.jar
Error: Could not find or load main class hi.jar
Caused by: java.lang.ClassNotFoundException: hi.jar
➜  snips scala -classpath hi.jar Main
hello, world

Output

Error: Could not find or load main class hi.jar

Expectation

Scala runner creates and runs an executable jar. It looks like it doesn't create META-INF/MANIFEST.MF.

[okarmusk]

I've got the same issue when run scala 3 from binary.
Bug is visible even if jar is build from sbt package where sbt is in version 1.5.2. Please check issue what has been added by @adpi2

[okarmusk]

I built the same source code with the same sbt version but I compile and run it with scala 2.13.6, it works.
Issue exists only in scala 3.

[som-snytt]

I'll take a swing at this, since it's such a useful niche feature. By swing, I mean a bunt down the 3rd base line.

[dwijnand]

Overlaps with #13081, so either make sure to coordinate with @BarkingBad.

[BarkingBad]

I tested it @anatoliykmetyuk and compiling files using scalac -d hi.jar hi.scala doesn't provide Manifest file. The way to get runnable jar is to run it by scala -save hi.scala and then run it as scala hi.jar

Should we introduce change to scalac that would make Manifest file?

[som-snytt]

Scala 2 distinguished

  -Xmain-class <path>              Class for manifest's Main-Class entry (only useful with -d <jar>)
  -Xscript <object>                Treat the source file as a script and wrap it in a main method.

and REPL

Other startup options:

 -i <file>    preload <file> before starting the REPL
 -I <file>    preload <file>, enforcing line-by-line interpretation
 -e <string>  execute <string> as if it were in a source file
 -save        save the compiled script in a jar for future use

[anatoliykmetyuk]

Should we introduce change to scalac that would make Manifest file?

I believe so. In any case this shouldn't be a -save option to produce a manifest for a runnable jar, as "save" isn't very related to "runnable".

[BarkingBad]

I guess bringing back -Xmain-class should be easy.
However, I don't know if -Xscript should be in the scope of this task and whether we need it at all in the compiler, we can run scripts by running scala <path-to-script>

[anatoliykmetyuk]

Yes, we don't need -Xscript anymore since we have a better, more seamless, scripting support in Scala 3.

[som-snytt]

I don't have a grid for Scala 3 runner options yet, let alone the recommended install procedure.

Probably -Xscript is not worth supporting, but with parser phase broken out, it could be supported third-party. It just wraps code in a main method in parser:

Using scala version 2.13.6 in this shell.
➜  snips cat simple.sc

println(s"hello, ${util.Properties.userName}")
➜  snips scalac -Xscript Main -d /tmp simple.sc
➜  snips scala -cp /tmp Main
hello, amarki
➜  snips scala -e 'println(s"hello, ${util.Properties.userName}")'
hello, amarki
➜  snips scala -howtorun:script simple.sc
hello, amarki

I don't know if the feature is used outside -e, or I guess people still use shell scripts, but there is some simplicity in avoiding REPL semantics.

[anatoliykmetyuk]

Yes, we discussed whether we should go Python-style or tolerate @main in scripts earlier. The argument for @main was not only simplicity but also the ability to accept command arguments in a natural way. Another argument was that scripts and programs ideally should be the same thing; we should strive to avoid differentiating between the two. Overall we got the consensus that @main in scripts is fine.