I have been trying to solve a memory issue in a Java program where we are loading an entire file into memory, base64 encoding it and then using it as a form parameter in a post request. This is cause OOME due to the extremely large file size.
I am working on a solution where I am able to stream the file through a base64 encoder, into the request body of an Http Post request. One of the common patterns I have noticed in all of the popular encoding libraries( Guava, java.util.Base64, android.util.Base64 and org.apache.batik.util ) is that if the library supports encoding with Streams, the Encoding is always done through an OutputStream and the Decoding is always done through an InputStream.
I am having trouble finding/determining the reasoning behind these decisions. Given that so many of these popular and well-written libraries align with this api design, I assume that there is a reason for this. It doesn't seem very difficult to adapt one of these decoders to become an InputStream or accept an InputStream, but I am wondering if there is a valid architectural reason these encoders are designed this way.
Why do common libraries do Base64 encoding through an OuputStream and Base64 decoding through an InputStream?
Examples to back up my claims:
java.util.Base64
- Base64.Decoder.wrap(InputStream stream)
- Base64.Encoder.wrap(OutputStream stream)
android.util.Base64
- Base64InputStream // An InputStream that does Base64 decoding on the data read through it.
- Base64OutputStream // An OutputStream that does Base64 encoding
google.common.io.BaseEncoding
- decodingStream(Reader reader)
- encodingStream(Writer writer)
org.apache.batik.util
- Base64DecodeStream implements InputStream
- Base64EncodeStream implements OutputStream
Well, yes, you can reverse it, but this makes the most sense. Base64 is used to make binary data - generated or operated on by the application - compatible with a text based outside environment. So the base 64 encoded data is always required on the outside and the decoded binary data is required on the inside.
An application generally doesn't perform any operations on the base 64 encoded data itself; it is just needed to communicate binary data with another application when a text interface is required or expected.
If you want to export your binary data to the outside, naturally you would use an output stream. If that data needs to be encoded in base 64, you make sure you send the data to an output stream that encodes to base 64.
If you want to import your binary data from the outside then you would use an input stream. If that data is encoded in base 64 then you first need to decode it, so you make sure you decode it before treating it as a binary stream.
Lets create a bit of a picture. Say you have an application that operates in a textual oriented environment but operates on binary data. The important part is the direction of the arrows from the context of the application on the left.
Then you get for the input (read calls):
{APPLICATION} <- (binary data decoding) <- (base64 decoding) <- (file input stream) <- [BASE 64 ENCODED FILE]
for this you naturally use input streams.
So let's look at the output (write calls):
{APPLICATION} -> (binary data encoding) -> (base64 encoding) -> (file output stream) -> [BASE 64 ENCODED FILE]
for this you naturally use output streams.
These stream can be connected to each other by chaining them together, i.e. using one stream as parent of the other stream.
Here is an example in Java. Note that creating the binary encoder/decoder in the data class itself is a bit ugly; generally you would use another class for that - I hope it suffices for demonstration purposes.
import static java.nio.charset.StandardCharsets.UTF_8;
import java.io.DataInputStream;
import java.io.DataOutputStream;
import java.io.FileInputStream;
import java.io.FileNotFoundException;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.util.Base64;
public class BinaryHandlingApplication {
/**
* A data class that encodes to binary output, e.g. to interact with an application in another language.
*
* Binary format: [32 bit int element string size][UTF-8 element string][32 bit element count]
* The integers are signed, big endian values.
* The UTF-8 string should not contain a BOM.
* Note that this class doesn't know anything about files or base 64 encoding.
*/
public static class DataClass {
private String element;
private int elementCount;
public DataClass(String element) {
this.element = element;
this.elementCount = 1;
}
public String getElement() {
return element;
}
public void setElementCount(int count) {
this.elementCount = count;
}
public int getElementCount() {
return elementCount;
}
public String toString() {
return String.format("%s count is %d", element, elementCount);
}
public void save(OutputStream out) throws IOException {
DataOutputStream dataOutputStream = new DataOutputStream(out);
// so here we have a chain of:
// a dataoutputstream on a base 64 encoding stream on a fileoutputstream
byte[] utf8EncodedString = element.getBytes(UTF_8);
dataOutputStream.writeInt(utf8EncodedString.length);
dataOutputStream.write(utf8EncodedString);
dataOutputStream.writeInt(elementCount);
}
public void load(InputStream in) throws IOException {
DataInputStream dataInputStream = new DataInputStream(in);
// so here we have a chain of:
// a datainputstream on a base 64 decoding stream on a fileinputstream
int utf8EncodedStringSize = dataInputStream.readInt();
byte[] utf8EncodedString = new byte[utf8EncodedStringSize];
dataInputStream.readFully(utf8EncodedString);
this.element = new String(utf8EncodedString, UTF_8);
this.elementCount = dataInputStream.readInt();
}
}
/**
* Create the a base 64 output stream to a file; the file is the text oriented
* environment.
*/
private static OutputStream createBase64OutputStreamToFile(String filename) throws FileNotFoundException {
FileOutputStream textOutputStream = new FileOutputStream(filename);
return Base64.getUrlEncoder().wrap(textOutputStream);
}
/**
* Create the a base 64 input stream from a file; the file is the text oriented
* environment.
*/
private static InputStream createBase64InputStreamFromFile(String filename) throws FileNotFoundException {
FileInputStream textInputStream = new FileInputStream(filename);
return Base64.getUrlDecoder().wrap(textInputStream);
}
public static void main(String[] args) throws IOException {
// this text file acts as the text oriented environment for which we need to encode
String filename = "apples.txt";
// create the initial class
DataClass instance = new DataClass("them apples");
System.out.println(instance);
// perform some operation on the data
int newElementCount = instance.getElementCount() + 2;
instance.setElementCount(newElementCount);
// write it away
try (OutputStream out = createBase64OutputStreamToFile(filename)) {
instance.save(out);
}
// read it into another instance, who cares
DataClass changedInstance = new DataClass("Uh yeah, forgot no-parameter constructor");
try (InputStream in = createBase64InputStreamFromFile(filename)) {
changedInstance.load(in);
}
System.out.println(changedInstance);
}
}
Especially note the chaining of the streams and of course the absence of any buffers whatsoever. I've used URL-safe base 64 (in case you want to use HTTP GET instead).
In your case, of course, you could generate a HTTP POST request using an URL and directly encode to the retrieved OutputStream stream by wrapping it. That way no base 64 encoded data needs to be (extensively) buffered. See examples on how to get to the OutputStream here.
Remember, if you need to buffer, you're doing it wrong.
As mentioned in the comments, HTTP POST doesn't need base 64 encoding but whatever, now you know how you can encode base 64 directly to a HTTP connection.
java.util.Base64 specific note:
Although base 64 is text, the base64 stream generates / consumes bytes;
it simply assumes ASCII encoding (this can be fun for UTF-16 text).
Personally I think this is a terrible design decision; they should have wrapped a Reader and Writer instead, even if that slows down encoding slightly.
To their defense, the various base 64 standards and RFC also get this wrong.