| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
this also demonstrates how to programmatically generate an extensive,
type-safe api for a stringly-typed interface
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This should allow for build to add other builds to their dependencies
and interact with them in a type-safe way. And ever regardless it seems
like good practice to never have the same class existing in the same
package or the top-level package even if they don’t end up on the same
classpath. This might also help make stack traces easier to understand.
Also improve error messages for mistakes with the build class, e.g.
constructor, super classes, etc.
|
|
|
|
| |
should be safe now that we rebuild based on lastModified
|
| |
|
|
|
|
| |
useful for `dotty run <file>`
|
|
|
|
|
| |
and prepares for allowing `run` and `runFlat` at
Dependency instead of Build level
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
using lastModified instead of a non-idempotent needsUpdate flag
this fixes a bug where dependees would not be rebuilt if cbt exited
or was killed after dependencies were already rebuilt.
|
| |
|
|
|
|
|
| |
this will make it possible to access lastModified times and cache them
in the following commits
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The concurrent hashmap approach to classloader caching was flawed.
Assume you have two concurrently running builds A and B and projects
P2 and P3 depending on project P1. And assume a time sequence where A
compiles P1, then compiles P2, then P1’s sources change, then B compiles
P1, then A compiles P3. At the end P2 and P3 will have different
versions of P1 as their parent classloaders. This is inconsistent.
The easiest way to work around this is making sure only one thread is
changing the classloader cache during it’s entire run. This would mean
either no concurrency or what we have done here, which is letting
threads work on a copy of the cache and replace the original cache in
the end using an atomic operation. This means the thread that finishes
last wins, but for caching that’s fine. Worst case some things aren’t
cached in a concurrent execution.
This change also means that we don’t need concurrent hashmaps for the
classloader cache anymore since no two theads will access the same
hashmap. We still need a concurrent hashmap for the class caches inside
of the classloaders as multiple threads can access the same
classloaders.
|
|
|
|
|
|
|
| |
This isn’t type-safe, but re-using that same hashmap for both keys and
classloaders allows to reduce the number of members in Context. Also
we can re-use the same hashMap for other things as well in the coming
commits, e.g. timestamps.
|
|
|
|
|
|
|
|
| |
Before it only affected jdk, because scala.Console captures our and err
before the swap.
This is needed when running main classes like Scaladoc or the compiler
and wanting to redirect output to standard error
|
| |
|
|
|
|
|
|
|
|
|
|
| |
A SecurityManager is once installed globally and stays the same across
all classloaders. TrapSecurityManager was installed this way, but it
looked up the `trapExitCode` in it's own classloader, while
classes in other classloaders (when cbt was using another version of
cbt) were writing it to their own classloader. This flag needs to be in
a global place instead to fix this, so we'll put it straight into the
TrapSecurityManager itself.
|
| |
|
|
|
|
| |
if multiple main classes are found
|
| |
|
|
|
|
| |
and work around the fact that the main method is not static (huh?)
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
Installing one globally for the JVM live-time and make behavior dependent
on a thread local variable seems safer than globally switching it out
and having race conditions.
Also now all other calls are forwarded to a potential Nailgun
SecurityManager, which should fix some bugs.
|
|
|
|
|
|
|
| |
Proxy settings can be supplied either via Java system-properties or via
environment variables (http_proxy/https_proxy/no_proxy).
Java system-properties take precedence over env vars. Evaluation of
proxy settings happens only during nailgun startup for now.
|
|
|
|
| |
changes builds from using immutable.Seq to using Predef.Seq in favor of simplicity over purity. This has been discussed on gitter.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
One large commit, because it is was hard to do these things in isolation or to separate them now.
CBT now knows how to load other versions of itself
- Support for reproducible builds (!), by providing a CBT git URL and hash to tie build to
- Support for composing builds using different CBT versions (!)
- introduce (in compatibility/) Java interfaces all CBT versions need to stay compatible with, so they can talk to each other. And put extension methods to these interfaces in cbt package object
Class loading
- add some sanity checks for class loading
- improve class loader invalidation to fix bugs
- implement caching in Java land class loaders. In particular to prevent the system class loader to repeatedly generate ClassNotFound exceptions in each sink of the class loader DAG for non JDK classes (meaning major speed up for projects with many classes).
- getting rid of transient class loader cache unifying into "persistent" one instead (which is still wrong as invalidation eventually needs to invalidate entire sub graphs of the class loading DAG, not single class loaders. Seems like we'll have to abandon the hashmap based approach and tie caching to dependency objects)
Other Caching
- cache dependencies extracted from xml files, which was one major time killer, but invalidate cache when cbt changed (maven dependency user facing api needs simplification now!)
- memorize last successful compile time in the file system rather than memory, to guard against unnecessary recompiling even across launches (or when using cbt direct)
Structural improvements
- Factor out ClassLoaderCache on Java land into its own class.
- Port MultiClassLoader to Java land, to better compose classloaders in NailgunLauncher.
- Remove many global constants and variables (in object paths and in NailgunLauncher) and pass them through instead. Needed for composing of builds.
- move more code from resolver into Lib for less entanglement with classes (needed to compatibility interfaces) and better re-usability
- remove canBeCached. Everything can be cached now, but we need to be careful about correct invalidation.
- remove build announcing produced jars. We can add if ever needed.
- change callNullary to return exit code instead of Unit as preparation for next commit introducing "recursive"
ScalaTest
- Makes ScalaTest support work (still a bit too inflexible, but mostly works well)
|
| |
|
| |
|
|
|
|
|
| |
Adds support for cross building for multiple scala versions
And for automatically appending -SNAPSHOT to the id, when publishing a snapshot.
|
|
|
|
| |
in case of a BuildBuild.
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit also
- requires Dependencies to explicitly implement canBeCached
- unifies some logic for dependency downloading
- moves SBT-like dependency DSL into its own trait
- error message showing build directory for exceptions in builds
Not splitting this up in favor of faster progress to 1.0.
The user facing API can probably be slightly improved using implicits
inside of BasicBuild, but we can do that later.
|
| |
|
|
|
|
| |
This seem desirable, but also fixes a test failure for which I have no idea why it didn't fail before but now.
|
|
|
|
|
|
| |
(and safe up to 0.1s for each call)
There still seem to be 2 bugs related to CBT development in the code. One if you simpy save a stage1 file unchanged and re-run cbt, it fails to call Stage2.run reflectively. Also in case of compile errors in stage1, a TrappedExitCode exception is thrown and not caught.
|