mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-11-30 08:12:05 +00:00
aad605e498
Differential Revision: https://phabricator.services.mozilla.com/D158643
255 lines
7.9 KiB
JavaScript
255 lines
7.9 KiB
JavaScript
/* This Source Code Form is subject to the terms of the Mozilla Public
|
|
* License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
|
|
|
"use strict";
|
|
|
|
const DevToolsUtils = require("resource://devtools/shared/DevToolsUtils.js");
|
|
const { dumpv } = DevToolsUtils;
|
|
const EventEmitter = require("resource://devtools/shared/event-emitter.js");
|
|
|
|
DevToolsUtils.defineLazyGetter(this, "IOUtil", () => {
|
|
return Cc["@mozilla.org/io-util;1"].getService(Ci.nsIIOUtil);
|
|
});
|
|
|
|
DevToolsUtils.defineLazyGetter(this, "ScriptableInputStream", () => {
|
|
return Components.Constructor(
|
|
"@mozilla.org/scriptableinputstream;1",
|
|
"nsIScriptableInputStream",
|
|
"init"
|
|
);
|
|
});
|
|
|
|
const BUFFER_SIZE = 0x8000;
|
|
|
|
/**
|
|
* This helper function (and its companion object) are used by bulk senders and
|
|
* receivers to read and write data in and out of other streams. Functions that
|
|
* make use of this tool are passed to callers when it is time to read or write
|
|
* bulk data. It is highly recommended to use these copier functions instead of
|
|
* the stream directly because the copier enforces the agreed upon length.
|
|
* Since bulk mode reuses an existing stream, the sender and receiver must write
|
|
* and read exactly the agreed upon amount of data, or else the entire transport
|
|
* will be left in a invalid state. Additionally, other methods of stream
|
|
* copying (such as NetUtil.asyncCopy) close the streams involved, which would
|
|
* terminate the debugging transport, and so it is avoided here.
|
|
*
|
|
* Overall, this *works*, but clearly the optimal solution would be able to just
|
|
* use the streams directly. If it were possible to fully implement
|
|
* nsIInputStream / nsIOutputStream in JS, wrapper streams could be created to
|
|
* enforce the length and avoid closing, and consumers could use familiar stream
|
|
* utilities like NetUtil.asyncCopy.
|
|
*
|
|
* The function takes two async streams and copies a precise number of bytes
|
|
* from one to the other. Copying begins immediately, but may complete at some
|
|
* future time depending on data size. Use the returned promise to know when
|
|
* it's complete.
|
|
*
|
|
* @param input nsIAsyncInputStream
|
|
* The stream to copy from.
|
|
* @param output nsIAsyncOutputStream
|
|
* The stream to copy to.
|
|
* @param length Integer
|
|
* The amount of data that needs to be copied.
|
|
* @return Promise
|
|
* The promise is resolved when copying completes or rejected if any
|
|
* (unexpected) errors occur.
|
|
*/
|
|
function copyStream(input, output, length) {
|
|
const copier = new StreamCopier(input, output, length);
|
|
return copier.copy();
|
|
}
|
|
|
|
function StreamCopier(input, output, length) {
|
|
EventEmitter.decorate(this);
|
|
this._id = StreamCopier._nextId++;
|
|
this.input = input;
|
|
// Save off the base output stream, since we know it's async as we've required
|
|
this.baseAsyncOutput = output;
|
|
if (IOUtil.outputStreamIsBuffered(output)) {
|
|
this.output = output;
|
|
} else {
|
|
this.output = Cc[
|
|
"@mozilla.org/network/buffered-output-stream;1"
|
|
].createInstance(Ci.nsIBufferedOutputStream);
|
|
this.output.init(output, BUFFER_SIZE);
|
|
}
|
|
this._length = length;
|
|
this._amountLeft = length;
|
|
let _resolve;
|
|
let _reject;
|
|
this._deferred = new Promise((resolve, reject) => {
|
|
_resolve = resolve;
|
|
_reject = reject;
|
|
});
|
|
this._deferred.resolve = _resolve;
|
|
this._deferred.reject = _reject;
|
|
|
|
this._copy = this._copy.bind(this);
|
|
this._flush = this._flush.bind(this);
|
|
this._destroy = this._destroy.bind(this);
|
|
|
|
// Copy promise's then method up to this object.
|
|
// Allows the copier to offer a promise interface for the simple succeed or
|
|
// fail scenarios, but also emit events (due to the EventEmitter) for other
|
|
// states, like progress.
|
|
this.then = this._deferred.then.bind(this._deferred);
|
|
this.then(this._destroy, this._destroy);
|
|
|
|
// Stream ready callback starts as |_copy|, but may switch to |_flush| at end
|
|
// if flushing would block the output stream.
|
|
this._streamReadyCallback = this._copy;
|
|
}
|
|
StreamCopier._nextId = 0;
|
|
|
|
StreamCopier.prototype = {
|
|
copy() {
|
|
// Dispatch to the next tick so that it's possible to attach a progress
|
|
// event listener, even for extremely fast copies (like when testing).
|
|
Services.tm.dispatchToMainThread(() => {
|
|
try {
|
|
this._copy();
|
|
} catch (e) {
|
|
this._deferred.reject(e);
|
|
}
|
|
});
|
|
return this;
|
|
},
|
|
|
|
_copy() {
|
|
const bytesAvailable = this.input.available();
|
|
const amountToCopy = Math.min(bytesAvailable, this._amountLeft);
|
|
this._debug("Trying to copy: " + amountToCopy);
|
|
|
|
let bytesCopied;
|
|
try {
|
|
bytesCopied = this.output.writeFrom(this.input, amountToCopy);
|
|
} catch (e) {
|
|
if (e.result == Cr.NS_BASE_STREAM_WOULD_BLOCK) {
|
|
this._debug("Base stream would block, will retry");
|
|
this._debug("Waiting for output stream");
|
|
this.baseAsyncOutput.asyncWait(this, 0, 0, Services.tm.currentThread);
|
|
return;
|
|
}
|
|
throw e;
|
|
}
|
|
|
|
this._amountLeft -= bytesCopied;
|
|
this._debug("Copied: " + bytesCopied + ", Left: " + this._amountLeft);
|
|
this._emitProgress();
|
|
|
|
if (this._amountLeft === 0) {
|
|
this._debug("Copy done!");
|
|
this._flush();
|
|
return;
|
|
}
|
|
|
|
this._debug("Waiting for input stream");
|
|
this.input.asyncWait(this, 0, 0, Services.tm.currentThread);
|
|
},
|
|
|
|
_emitProgress() {
|
|
this.emit("progress", {
|
|
bytesSent: this._length - this._amountLeft,
|
|
totalBytes: this._length,
|
|
});
|
|
},
|
|
|
|
_flush() {
|
|
try {
|
|
this.output.flush();
|
|
} catch (e) {
|
|
if (
|
|
e.result == Cr.NS_BASE_STREAM_WOULD_BLOCK ||
|
|
e.result == Cr.NS_ERROR_FAILURE
|
|
) {
|
|
this._debug("Flush would block, will retry");
|
|
this._streamReadyCallback = this._flush;
|
|
this._debug("Waiting for output stream");
|
|
this.baseAsyncOutput.asyncWait(this, 0, 0, Services.tm.currentThread);
|
|
return;
|
|
}
|
|
throw e;
|
|
}
|
|
this._deferred.resolve();
|
|
},
|
|
|
|
_destroy() {
|
|
this._destroy = null;
|
|
this._copy = null;
|
|
this._flush = null;
|
|
this.input = null;
|
|
this.output = null;
|
|
},
|
|
|
|
// nsIInputStreamCallback
|
|
onInputStreamReady() {
|
|
this._streamReadyCallback();
|
|
},
|
|
|
|
// nsIOutputStreamCallback
|
|
onOutputStreamReady() {
|
|
this._streamReadyCallback();
|
|
},
|
|
|
|
_debug(msg) {
|
|
// Prefix logs with the copier ID, which makes logs much easier to
|
|
// understand when several copiers are running simultaneously
|
|
dumpv("Copier: " + this._id + " " + msg);
|
|
},
|
|
};
|
|
|
|
/**
|
|
* Read from a stream, one byte at a time, up to the next |delimiter|
|
|
* character, but stopping if we've read |count| without finding it. Reading
|
|
* also terminates early if there are less than |count| bytes available on the
|
|
* stream. In that case, we only read as many bytes as the stream currently has
|
|
* to offer.
|
|
* TODO: This implementation could be removed if bug 984651 is fixed, which
|
|
* provides a native version of the same idea.
|
|
* @param stream nsIInputStream
|
|
* The input stream to read from.
|
|
* @param delimiter string
|
|
* The character we're trying to find.
|
|
* @param count integer
|
|
* The max number of characters to read while searching.
|
|
* @return string
|
|
* The data collected. If the delimiter was found, this string will
|
|
* end with it.
|
|
*/
|
|
function delimitedRead(stream, delimiter, count) {
|
|
dumpv(
|
|
"Starting delimited read for " + delimiter + " up to " + count + " bytes"
|
|
);
|
|
|
|
let scriptableStream;
|
|
if (stream instanceof Ci.nsIScriptableInputStream) {
|
|
scriptableStream = stream;
|
|
} else {
|
|
scriptableStream = new ScriptableInputStream(stream);
|
|
}
|
|
|
|
let data = "";
|
|
|
|
// Don't exceed what's available on the stream
|
|
count = Math.min(count, stream.available());
|
|
|
|
if (count <= 0) {
|
|
return data;
|
|
}
|
|
|
|
let char;
|
|
while (char !== delimiter && count > 0) {
|
|
char = scriptableStream.readBytes(1);
|
|
count--;
|
|
data += char;
|
|
}
|
|
|
|
return data;
|
|
}
|
|
|
|
module.exports = {
|
|
copyStream,
|
|
delimitedRead,
|
|
};
|