On Tue, 24 Sep 2024 13:31:25 GMT, Eirik Bjørsnøs <eir...@openjdk.org> wrote:
> Please consider this PR which makes `ZipFileInflaterInputStream` release its > `Inflater` instance back to the pool when observing that the stream has been > fully consumed. > > This is motivated by the following observations: > > * After a `ZipFileInflaterInputStream` has reached the end of stream, its > `Inflater` instance no longer does useful work. > * Obtaining an input stream via `ZipFile::getInputStream` without properly > closing it is probably common. Obtaining it without fully consuming it is > probably more rare. (Something like > `classLoader.getInputStream(entry).readAllBytes()` is probably common) > * While GC will eventually release the `Inflater` when the `Cleaner` closes > the stream, this will only happen at some later point in time. In the > meantime, `ZipFile::getInputStream` may produce many new `Inflater` > instances. These will all be released to the pool once GC eventually catches > up. > * Once an `Inflater` is released to to pool, it will stay there for the > lifetime of the `ZipFile` > * The lifetime of a `ZipFile` may often be as long as the application > (consider class loaders). > > This PR suggests the following changes: > > * Rename the existing field `ZipFileInflaterInputStream.eof` to > `compressedEof`. (This tracks the EOF of the filtered input stream) > * Add a new field `ZipFileInflaterInputStream.eof` to track the EOF of > uncompressed data. > * Override `ZipFileInflaterInputStream.read(byte[] b, int off, int len)` to > detect the EOF of uncompressed data and release the `Inflater` instance back > to the pool when it is no longer needed. > * To protect the `Inflater` instance from being used after being released and > reset, the following updates are needed: > * `ZipFileInflaterInputStream.read` must check if EOF has been reached but > that the stream has not yet been closed. In this case, EOF should be returned. > * Similarly, `ZipFileInflaterInputStream.available` needs an update to > return 0 if EOF has been detected. > > The PR adds a test `ReleaseInflaterOnEOF` which verifies that all fully > consumed input streams are backed by the same `Inflater` instance. > > This PR was inspired by JDK-7031076 and this discussion: > https://mail.openjdk.org/pipermail/core-libs-dev/2011-March/006341.html This pull request has been closed without being integrated. ------------- PR: https://git.openjdk.org/jdk/pull/21157
