New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
The alternative, flat representation of classpath elements #4176
Commits on Nov 28, 2014
-
Abstract over ClassPath and ClassRep
This commit is intended to create the possibility to plug in into the compiler an alternative classpath representation which would be possibly more efficient, use less memory etc. Such an implementation - at least at the beginning - should exist next to the currently existing one and be possible to turn on using a flag. Several places in the compiler have a direct dependency on the classpath implementation. Examples include backend's icode generator and reader, SymbolLoaders, ClassfileParser. After closer inspection, one realizes that all those places depend only on a very small subset of classpath logic: they need to lookup classes from classpath. Hence there's introduced ClassFileLookup trait that encapsulates that functionality. The ClassPath extends that trait and an alternative one also must do it. There's also added ClassRepresentation - the base trait for ClassRep (the inner class of ClassPath). Thanks to that the compiler uses a type which is not directly related to the particular classpath representation as it was doing until now.
Commits on Nov 30, 2014
-
Use new asClassPathString method and create FileUtils for classpath
The method asClasspathString is now deprecated. Moreover it's moved to ClassFileLookup in the case someone was using it in some project (an alternative classpath also will support it - just in the case). All its usages existing in Scala sources are changed to asClassPathString method. The only difference is the name. Some operations on files or their names are moved from ClassPath to the newly created FileUtils dedicated to classpath. It will be possible to reuse them when implementing an alternative classpath representation. Moreover such allocation-free extension methods like the one added in this commit will improve the readability.
-
Define interface for flat classpath and add package loader using it
This commit introduces the base trait for flat classpath - an alternative classpath representation. In accordance with the idea and the experimental implementation of @gkossakowski, this representation will try to make the best use of the specificity of a given file type instead of using AbstractFile everywhere. It's possible as .NET backend is no longer supported and we can focus on Java-specific types of files. FlatClassPath extends ClassFileLookup which provides the common interface used also by existing ClassPath. The new implementation is called flat because it's possible to query the whole classpath using just single instance. In the case of the old (recursive) representation there's the structure of nested classpath objects, where each such an object can return only entries from one level of hierarchy but it returns also another classpath objects for nested levels included in it. That's why there's added dedicated PackageLoaderUsingFlatClassPath in SymbolLoaders - approaches are different so also the way of loading packages has to be different. The new package loader is currently unused. There's added also PackageNameUtils which will provide common methods used by classpath implementations for various file types.
-
Add the flat classpath type aggregating flat classpath instances
There's added AggregateFlatClassPath - an equivalent of MergedClassPath from the old implementation. It is supposed to group classpath instances handling different files being directories, zips or jars. Unlike in the case of the old (recursive) implementation, there won't be a deep, nested hierarchy of classpath instances - just one root (aggregate) and a flat structure of its children. AggregateFlatClassPath ensures the distinction of classpath entries and merges corresponding entries for class and source files into one entry. This is required as SymbolLoaders class makes use of this kind of ClassRepresentation. There are also added unit tests which check whether AggregateFlatClassPath obtains correct entries from classpath instances specified in a constructor and whether it preserves the ordering in the case of repeated entries. There's added a test type of flat classpath using VirtualFiles so it's easy to check the real behaviour.
-
Add flat classpath implementation for directories
There's added the flat classpath implementation for directories using java.util.File directly. Since we work with a real directory - not the AbstractFile - we don't need to iterate all entries of a file to get inner entries of some package. We can just find an adequate directory for a package. There are added implementations for a class- and a sourcepath. Both extend DirectoryFileLookup which provides common logic.
-
Add flat classpath implementation for zip and jar files
This commit adds an implementation of flat classpath which can handle both jar and vanilla zip files. In fact there are two versions - for a class- and a sourcepath. Both extend ZipArchiveFileLookup which provides common logic. They use FileZipArchive. @gkossakowski made a comparison of different ways of handling zips and jars (e.g. using javac's ZipFileIndex). He stated that general efficiency of FileZipArchive, taking into account various parameters, is the best. FileZipArchive is slightly changed. From now it allows to find the entry for directory in all directory entries without iterating all entries regardless of a type. Thanks to that we can simply find a directory for a package - like in the case of DirectoryFileLookup. There's also added possibility to cache classpath representation of classpath elements from jar and zip files across compiler instances. The cache is just a map AbstractFile -> FlatClassPath. It should reduce the number of created classpath and file instances e.g. in the case of many ScalaPresentationCompilers in Scala IDE. To prevent the possibility to avoid a cache, caches are created as a part of factories responsible for the creation of these types of the flat classpath.
-
Add flat classpath implementation using ManifestResources
There's added the flat classpath type using ManifestResources, closely related to the support for JSR-223 (Scripting for the Java Platform). It uses classes listed in the manifest file placed in the JAR. It's related to jar files so it's created using ZipAndJarFlatClassPathFactory and is cached. In general currently it's not possible to use it in Scala out of the box (without using additional tools such as jarlister) as this support is postponed. The old classpath has been properly prepared in the PR created by @rjolly scala#2238 so the new one also got this feature. ManifestResources is a ZipArchive without a real underlying file placed on a disk and in addition implementing some methods declared in AbstractFile as unsupported operations. Therefore the implementation has to use the iterator. I wanted to have the similar behaviour as in the case of directories and zip/jar files - be able to get a directory entry for a package without iterating all entries. This is achieved by iterating all entries only once and caching packages. This flat classpath type was the last needed one.
-
Create base classpath factory and an implementation for the flat cp
The part of the functionality of a ClassPathContext has been moved to the base trait ClassPathFactory so it can be reused by the newly created FlatClassPathFactory. This new implementation works in similar manner as the ClassPathContext with this difference that it just creates instances of flat classpath. This change doesn't modify the behaviour of the compiler as the interface and the way ClassPathContext works didn't change. Moreover FlatClassPathFactory is currently unused.
-
Create dedicated path resolver for the flat classpath representation
This commit adds dedicated FlatClassPathResolver loading classpath entries as FlatClassPath. Most of the common logic from PathResolver for the old classpath has been moved to the base, separate class which isn't dependent on a particular classpath representation. Thanks to that it was possible to reuse it when creating an adequate path resolver for the flat classpath representation. This change doesn't modify the way the compiler works. It also doesn't change nothing from the perspective of someone who already uses PathResolver in some project or even extends it - at least as long as he/she doesn't need to use flat classpath. There are also added JUnit tests inter alia comparing entries created using the old and the new classpath representations (whether the flat one created using the new path resolver returns the same entries as the recursive one).
Commits on Dec 1, 2014
-
The structure of scalap's Main has been refactored. EmptyClasspath is deleted. It looks that it was unused since this commit: scala@e594fe5 Also classpath logging is changed and now uses asClassPathString method. It was needed to modify one test because of that but it won't depend on a particular representation. There aren't changes in the way scalap works.
Commits on Dec 5, 2014
-
Integrate flat classpath with the compiler
This commit integrates with the compiler the whole flat classpath representation build next to the recursive one as an alternative. From now flat classpath really works and can be turned on. There's added flag -YclasspathImpl with two options: recursive (the default one) and flat. It was needed to make the dynamic dispatch to the particular classpath representation according to the chosen type of a classpath representation. There's added PathResolverFactory which is used instead of a concrete implementation of a path resolver. It turned out that only a small subset of path resolvers methods is used outside this class in Scala sources. Therefore, PathResolverFactory returns an instance of a base interface PathResolverResult providing only these used methods. PathResolverFactory in combination with matches in some other places ensures that in all places using classpath we create/get the proper representation. Also the classPath method in Global is modified to use the dynamic dispatch. This is very important change as a return type changed to the base ClassFileLookup providing subset of old ClassPath public methods. It can be problematic if someone was using in his project the explicit ClassPath type or public methods which are not provided via ClassFileLookup. I tested flat classpath with sbt and Scala IDE and there were no problems. Also was looking at sources of some other projects like e.g. Scala plugin for IntelliJ and there shouldn't be problems, I think, but it would be better to check these changes using the community build. Scalap's Main.scala is changed to be able to use both implementations and also to use flags related to the classpath implementation. The classpath invalidation is modified to work properly with the old (recursive) classpath representation after changes made in a Global. In the case of the attempt to use the invalidation for the flat cp it just throws exception with a message that the flat one currently doesn't support the invalidation. And also that's why the partest's test for the invalidation has been changed to use (always) the old implementation. There's added an adequate comment with TODO to this file. There's added partest test generating various dependencies (directories, zips and jars with sources and class files) and testing whether the compilation and further running an application works correctly, when there are these various types of entries specified as -classpath and -sourcepath. It should be a good approximation of real use cases.
-
Create possibility to skip flat classpath caching
There's added -YdisableFlatCpCaching option to ScalaSettings which allows user to disable caching of flat representation of classpath elements.
-
Cleanup and refactoring - semicolons, unused or commented out code
This commit contains some minor changes made by the way when implementing flat classpath. Sample JUnit test that shows that all pieces of JUnit infrastructure work correctly now uses assert method form JUnit as it should do from the beginning. I removed commented out lines which were obvious to me. In the case of less obvious commented out lines I added TODOs as someone should look at such places some day and clean them up. I removed also some unnecessary semicolons and unused imports. Many string concatenations using + have been changed to string interpolation. There's removed unused, private walkIterator method from ZipArchive. It seems that it was unused since this commit: scala@9d4994b However, I had to add an exception for the compatibility checker because it was complaining about this change. I made some trivial corrections/optimisations like use 'findClassFile' method instead of 'findClass' in combination with 'binary' to find the class file.
-
Add benchmarks to compare recursive and flat cp representations
The goal of these changes is to add possibility to: - compare an efficiency and a content of both cp implementations (ClassPathImplComparator) - examine the memory consumption by creating a lot of globals using a specified classpath (ClassPathMemoryConsumptionTester) - it can be considered as e.g. some approximation of ScalaPresentationCompilers in Scala IDE when working with many projects ClassPathMemoryConsumptionTester is placed in main (I mean not test) sources so thanks to that it has properly, out of the box configured boot classpath etc. and it's easy to use it, e.g.: scala scala.tools.nsc.ClassPathMemoryConsumptionTester -YclasspathImpl:<implementation_to_test> -cp <some_cp> -sourcepath <some_sp> -requiredInstances 50 SomeFileToCompile.scala At the end it waits for the "exit" command so there can be used some profiler like JProfiler to look how the given implementation behaves. Also flat classpath implementation is set as a default one to test it on Jenkins. This particular change must be reverted when all tests will pass because for now it's not desirable to make it permanently the default representation.
-
Turn off flat classpath by default, mark one of its classes as sealed
This commit addresses code review comments. The flat classpath is no longer the default classpath representation. It was the default one just for the test purposes. For now it's not desirable to make it permanently the default representation. ZipAndJarFileLookupFactory is marked as sealed - it should help to limit the ways of creating flat classpath instances for zips and jars.