| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
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.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
cbtHasChanged does never change across one run. However
classLoaderRecursion removes the corresponding classloader if
needsUpdate is true. It is called multiple times meaning it removes
and re-creates the classloader on every call leading to potentially
multiple conflicting cbt classloaders. needsUpdate should never be
returning true more than ones across a single run. Otherwise it can
lead to this error:
java.lang.LinkageError: loader constraint violation in interface itable initialization: when resolving method "Build.lib()Lcbt/Lib;" the class loader (instance of cbt/URLClassLoader) of the current class, Build, and the class loader (instance of cbt/URLClassLoader) for interface cbt/BaseBuild have different Class objects for the type cbt/Lib used in the signature
|
| |
|
|
|
| |
and it becomes slightly easier to maintain source compatibility
between cbt versions
|
| |
|
|
| |
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)
|
| | |
|
| | |
|
| |
|
|
| |
and doing less string processing
|
| | |
|
| | |
|
| |
|
|
| |
to avoid creating classloaders for the same files more than once, which leads to conflicts.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
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 fixes the issues with crashes loading stage2 introduces in 57de43907e05d4cd3986e2994e0e3bff93e09b4e. Also should be faster when using nailgun.
|
| |
|
|
| |
structurally. (didn't cause any problem right now, but could some day)
|
| |
|
|
| |
tag contents
|
| |
|
|
|
|
| |
(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.
|
| |
|
|
|
|
|
|
|
| |
- Changed launcher to already load zinc
- use code generation to generate necessary dependencies
- changed resolver to linearize dependency DAG in a way that guarantees that every transitive dependee of a node in the DAG is a transitive dependee of that node in the linear sequence
- move exit code trapping code into java so it can be used for zinc early
There seems to be a bug in this version, where CBT crashes about half of the time with a "object is not an instance of declaring class" Exception during running the task from the build object via reflection.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
replace two level classloader (one for non-cachable dependencies with a cached parent one for cachable ones) with a hierachy of classloaders corresponding
this should eventually allow re-using CBT's classloader between stage1 and stage2
this change breaks the ScalaTest support for now
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
caching classloaders hierarchically without deadlocks
|
| | |
|
| |
|
|
| |
pair coded with @chavxo
|
| |
|
|
|
|
| |
properties and dependencyManagement pom inheritance features
pair-coded with @chavxo
|
| |
|
|
| |
determinism between dev machines and circle ci
|
| |\
| |
| |
| |
| |
| | |
# Conflicts:
# stage1/Stage1Lib.scala
# stage1/resolver.scala
|
| | | |
|
| | |\
| | |
| | | |
Improved how to express dependencies
|
| | | |
| | |
| | |
| | |
| | |
| | | |
Add
- Constructor syntax for cross-scala-version dependencies (as rightfully requested on reddit) and encouraged over SBT's still supported operator syntax
- Add support for classifiers other than "sources"
|
| | |/
| |
| |
| | |
now). Is CBT "reactive" now ;)?
|
| |/ |
|
| |
|
|
| |
logger as implicits
|
| | |
|
| |
|
|
| |
++ for strings everywhere. Added ++ method to File and URL and use it in many places
|
| |
|
