问题描述
我一直在尝试解决 Java 程序中的内存问题，我们将整个文件加载到内存中，对其进行 base64 编码，然后将其用作发布请求中的表单参数.这是由于文件太大而导致OOME.
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.
我正在研究一种解决方案，我可以通过 base64 编码器将文件流式传输到 Http Post 请求的请求正文中.我在所有流行的编码库(Guava、java.util.Base64、android.util.Base64 和 org.apache.batik.util)中注意到的常见模式之一是 if 库支持使用 Streams 进行编码，编码总是通过 OutputStream 完成，而解码总是通过 InputStream 完成.
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.
我很难找到/确定这些决定背后的原因.鉴于这些流行且编写良好的库中的许多都与此 api 设计保持一致，我认为这是有原因的.将这些解码器之一调整为 InputStream 或接受 InputStream 似乎并不难，但我想知道这些编码器以这种方式设计是否有有效的架构原因.
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.
为什么常见的库是通过 OuputStream 进行 Base64 编码，通过 InputStream 进行 Base64 解码?
Why do common libraries do Base64 encoding through an OuputStream and Base64 decoding through an InputStream?
支持我的声明的示例:
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
推荐答案
嗯，是的，你可以反转它，但这是最有意义的.Base64 用于使 二进制数据 - 由应用程序生成或操作 - 与基于文本的外部环境兼容.所以外部总是需要base 64编码的数据，内部需要解码的二进制数据.
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.
应用程序通常不会对 base 64 编码数据本身执行任何操作；它只需要与另一个应用程序进行二进制数据通信当需要或期望文本接口时.
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.
如果要将二进制数据导出到外部，自然会使用输出流.如果该数据需要以 base 64 编码，请确保将数据发送到编码为 base 64 的输出流.
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.
如果您想从外部导入二进制数据，那么您将使用输入流.如果该数据以 base 64 编码，则首先需要对其进行解码，因此请确保在将其视为二进制流之前对其进行解码.
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.
让我们创建一些图片.假设您有一个在面向文本的环境中运行但在二进制数据上运行的应用程序.重要的部分是左侧应用程序上下文中箭头的方向.
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.
然后你得到输入(读取调用):
Then you get for the input (read calls):
{APPLICATION} <- (binary data decoding) <- (base64 decoding) <- (file input stream) <- [BASE 64 ENCODED FILE]
为此，您自然会使用输入流.
for this you naturally use input streams.
那么让我们看看输出(写调用):
So let's look at the output (write calls):
{APPLICATION} -> (binary data encoding) -> (base64 encoding) -> (file output stream) -> [BASE 64 ENCODED FILE]
为此，您自然会使用输出流.
for this you naturally use output streams.
这些流可以通过将它们链接在一起相互连接，即使用一个流作为另一个流的父级.
These stream can be connected to each other by chaining them together, i.e. using one stream as parent of the other stream.
这是一个 Java 示例.请注意，在数据类本身中创建二进制编码器/解码器有点难看；通常你会为此使用另一个类 - 我希望它足以用于演示目的.
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);
}
}
特别注意流的链接，当然也没有任何缓冲区随便.我使用了 URL 安全的 base 64(如果您想使用 HTTP GET 代替).
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).
当然，在您的情况下，您可以使用 URL 生成 HTTP POST 请求，并直接编码到检索到的 OutputStream 流，方法是包装它.这样就不需要(广泛)缓冲 base 64 编码数据.查看有关如何访问 OutputStream 此处的示例.
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.
记住，如果你需要缓冲，那你就错了.
正如评论中提到的，HTTP POST 不需要 base 64 编码，但无论如何，现在您知道如何将 base 64 直接编码为 HTTP 连接了.
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 具体说明:虽然 base 64 是文本，但 base64 流生成/消耗字节；它只是假设 ASCII 编码(这对于 UTF-16 文本可能很有趣).我个人认为这是一个糟糕的设计决定.他们应该包装一个 Reader 和 Writer 代替，即使这会稍微减慢编码速度.
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.
为了他们的辩护，各种 base 64 标准和 RFC 也犯了这个错误.
To their defense, the various base 64 standards and RFC also get this wrong.