RFR: 8341666: FileInputStream doesn't support readAllBytes() or readNBytes(int) on pseudo devices

[Alan Bateman]

On Thu, 24 Oct 2024 11:29:01 GMT, Jaikiran Pai <jpai at openjdk.org> wrote:

>> Modify `FileInputStream` (FIS) to fall back to the superclass implementations of `readAllBytes()`, `readNBytes(int)`, `skip()`, and `transferTo` when the input source would otherwise fail with "Illegal Seek" in the FIS implementation, such as for the standard input or a named pipe.
>
> src/java.base/share/classes/java/io/FileInputStream.java line 334:
> 
>> 332:     @Override
>> 333:     public byte[] readAllBytes() throws IOException {
>> 334:         if (!canSeek())
> 
> Since `canSeek()` is implemented as a native call whose result depends on the file descriptor of this `FileInputStream`, do you think we should reduce these native calls in this change and call `canSeek()` just once (and store that state) in the constructors of this class (of course, after the call to open())?
> If we do that, we might have to consider whether the `FileInputStream(FileDescriptor fdObj)` will end up throwing an exception (from the native canSeek() implementation) if the `FileDescriptor` passed to the constructor is not `valid()` https://docs.oracle.com/en/java/javase/23/docs/api/java.base/java/io/FileDescriptor.html#valid()

readAllBytes reads to EOF so there isn't any benefit to caching. I think the main thing with this PR is whether canSeek is the right thing to use.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/21673#discussion_r1815043405