Cut dependencies on future modules #5830
Conversation
|
The remaining test failures look like this: The osgi tests all pass if this access to --- a/src/reflect/scala/reflect/internal/SymbolTable.scala
+++ b/src/reflect/scala/reflect/internal/SymbolTable.scala
@@ -138,7 +138,8 @@ abstract class SymbolTable extends macros.Universe
/** Dump each symbol to stdout after shutdown.
*/
- final val traceSymbolActivity = System.getProperty("scalac.debug.syms") != null
+ final val traceSymbolActivity = sys.props contains "scalac.debug.syms"
object traceSymbols extends {
val global: SymbolTable.this.type = SymbolTable.this
} with util.TraceSymbolActivity |
|
I'm looking into this one, I think I'm getting somewhere. |
|
Thanks! Re: osgi, does this have something to do with osgi enforcing boundaries and somehow the dependency on scala.sys was inferred and not explicitly specified? |
|
No, I don't think it has to do with osgi itself, but with the classloader type used in runtime reflection. It seems to be specific to package objects. I'll figure it out I think. |
Na klar! |
|
Finally figured it out... Your thorough work removed each and every dependency from scala-reflect to |
|
Now this is really weird: the $ cat Test.scala
object Test {
def main(args: Array[String]) {
new Thread(new Runnable {
def run = { Thread.sleep(1000); println("hi") }
}).start
}
}
$ scalac12 Test.scala
$ time scala12 Test.scala
hi
/Users/luc/scala/scala-2.12/bin/scala Test.scala 0.77s user 0.09s system 42% cpu 2.042 total
$ time ../build/quick/bin/scala Test
../build/quick/bin/scala Test 0.68s user 0.13s system 103% cpu 0.780 total
$[edit]: not so weird after all, see comment below: #5830 (review) |
| @@ -101,5 +101,5 @@ class MainGenericRunner { | |||
| } | |||
|
|
|||
| object MainGenericRunner extends MainGenericRunner { | |||
| def main(args: Array[String]): Unit = if (!process(args)) sys.exit(1) | |||
| def main(args: Array[String]): Unit = System.exit(if (process(args)) 0 else 1) | |||
There was a problem hiding this comment.
This changes behavior.. if process is true, we should not call exit. Otherwise this will kill the application without waiting for spawned non-daemon threads to finish. https://github.com/scala/scala/blob/v2.12.1/test/files/run/lazy-concurrent.scala is an example.
|
Green now! |
|
I'll squash some of the smaller commits into one to make them fewer. |
lrytz
changed the title
Since we no longer want to depend on `sys.error`, the compiler
also shouldn't generate calls to it.
This also fixes the OSGI tests. The failure was interesting: once
there were no more refernces to `scala.sys` left in scala-reflect,
the dependency on that package disappeared from scala-reflect.jar's
MANIFEST.MF. When instantiating scala.reflect.runtime.universe through
an osgi classloader, `getPackageObject("scala.sys")` in definitions
failed becasue the package could not be found through that classloader.
`scala.concurrent`, `scala.Console`, `scala.compat._`, `sys.runtime`, `sys.addShutdownHook`, `sys.allThreads`, `sys.env`. Also fixes a test reducing references to static modules (Console, Predef) delays initialization of object `NoStackTrace`. In `t2318`, this may now happen *after* the security manager is installed. Therefore the security manager needs to allow reading the system property `scala.control.noTraceSuppression`
Use `throw new RuntimeException` (or a subclass) instead.
In scala/scala#5830, the definition of `Sys_error` was removed from `Definitions.scala`. This commit reintroduces it in our own codebase, as a quickfix to scala-js#2880. This will need to be revisited as part of scala-js#2876 not use `sys.error` at all, but that is less urgent.
In scala/scala#5830, the definition of `Sys_error` was removed from `Definitions.scala`. This commit reintroduces it in our own codebase, as a quickfix to scala-js#2880. This will need to be revisited as part of scala-js#2876 not use `sys.error` at all, but that is less urgent.
In scala/scala#5830, the definition of `Sys_error` was removed from `Definitions.scala`. This commit reintroduces it in our own codebase, as a quickfix to scala-js#2880. This will need to be revisited as part of scala-js#2876 not use `sys.error` at all, but that is less urgent.
Re-submission of #5677