EXPOSED-86 allowing blobs with streams of unknown size

[elektro-wolle]

InputStreams for large blobs or blobs with unknown size fail, see description in

https://youtrack.jetbrains.com/issue/EXPOSED-86/JdbcPreparedStatementImpl-fails-on-input-stream-with-unknown-size-breaks-CipherStreams-in-blobs.

This fix also skips the side-effect in the getter of ExposedBlob.bytes, which silently read the full input-stream and replacing it with an ByteArrayInputStream.

[joc-a]

I tested it and it appears that this line works for all cases. Why not use just this?

[elektro-wolle]

readAllBytes() will fail on contents larger than 2GB (maximum size of the ByteArray) and allocates the whole data within RAM. The setBinaryStream function transfers just small chunks to the database until the stream is fully read (and allocates only one of those small chunks on the heap).

[joc-a]

Thanks for clarifying.

One thing I'm concerned about is having to check the type of InputStream and do something specific for each type. InputStream has a very large number of sub-classes and I wonder how sustainable/maintainable it is to have a separate approach for each one instead of having a generic solution that could potentially work for most, if not all of them, if that's possible.

Also, I tried testing it using a 3GB file created using RandomAccessFile and it failed for all databases for being too large.

        val fileSize = 3L * 1024L * 1024L * 1024L // 3GB
        val filePath = "path_to_large_file.txt"

        try {
            val randomAccessFile = RandomAccessFile(filePath, "rw")
            randomAccessFile.setLength(fileSize)
            println("Large file created: $filePath")
            randomAccessFile.close()
        } catch (e: IOException) {
            e.printStackTrace()
        }
        
        val longBlob = ExposedBlob(
              inputStream = FileInputStream(filePath)
        )

Is your solution supposed to cover this case?

[elektro-wolle]

Unfortunately, there is no interface within JDK to decide if the InputStream has a known size. But maybe, it's easier to skip the functions with the provided length.

I'll have to look into your 3GB test-case, it should have worked, maybe, it is necessary to use Connection.createBlob instead of InputStream. I hoped, this would be transparently handled by the JDBC-driver.

[elektro-wolle]

I'll split the PR into two separate PRs: one for the available() bug (with is more urgent for me) and the blob handling with Connection.createBlob() (which currently fails on columns with default values).

[joc-a]

Thanks for your contribution. Left a comment.

[elektro-wolle]

Rebased onto current main and corrected title.