The place where all the data is stored for VFS parts loaded into a memory: name-ids, flags, user data, children.
The purpose is to avoid holding this data in separate immortal file/directory objects because that involves space overhead, significant
when there are hundreds of thousands of files.
The data is stored per-id in blocks of
SEGMENT_SIZE. File ids in one project tend to cluster together,
so the overhead for non-loaded id should not be large in most cases.
File objects are still created if needed. There might be several objects for the same file, so equals() should be used instead of ==.
The lifecycle of a file object is as follows:
1. The file has not been instantiated yet, so
getFileById(int, com.intellij.openapi.vfs.newvfs.impl.VirtualDirectoryImpl, boolean) returns null.
2. A file is explicitly requested by calling getChildren or findChild on its parent. The parent initializes all the necessary data (in a thread-safe context)
and creates the file instance. See
initFile(int, com.intellij.openapi.vfs.newvfs.impl.VfsData.Segment, int, java.lang.Object)
3. After that the file is live, an object representing it can be retrieved any time from its parent. File system roots are
kept on hard references in
PersistentFS
4. If a file is deleted (invalidated), then its data is not needed anymore, and should be removed. But this can only happen after
all the listener have been notified about the file deletion and have had their chance to look at the data the last time. See
killInvalidatedFiles()
5. The file with removed data is marked as "dead" (see
myDeadMarker, any access to it will throw
InvalidVirtualFileAccessException
Dead ids won't be reused in the same session of the IDE.