"Write after end": how to imitate gulp-watch with watchify?

Question

The following Gulp task does almost what I want.

const gulp = require('gulp');
const browserify = require('browserify');
const vinylStream = require('vinyl-source-stream');
const vinylBuffer = require('vinyl-buffer');
const watchify = require('watchify');
const glob = require('glob');
const jasmineBrowser = require('gulp-jasmine-browser');

gulp.task('test', function() {
    let testBundler = browserify({
        entries: glob.sync('src/**/*-test.js'),
        cache: {},
        packageCache: {},
    }).plugin(watchify);
    function updateSpecs() {
        return testBundler.bundle()
            .pipe(vinylStream(jsBundleName))
            .pipe(vinylBuffer())
            .pipe(jasmineBrowser.specRunner({console: true}))
            .pipe(jasmineBrowser.headless({driver: 'phantomjs'}));
    }
    testBundler.on('update', updateSpecs);
    updateSpecs();
});

It bundles all my Jasmine specs using Browserify and has them tested through gulp-jasmine-browser. It also watches all specs and all modules that they depend on and re-runs the tests if any of these modules changes.

The only ugly bit, which I'd really like to see solved, is that a new PhantomJS instance and a new Jasmine server are created every time updateSpecs is run. I was hoping to avoid that with code like the following:

gulp.task('test', function() {
    let testBundler = browserify({
        entries: glob.sync('src/**/*-test.js'),
        cache: {},
        packageCache: {},
    }).plugin(watchify);
    // persist the Jasmine server and PhantomJS browser
    let testServer = jasmineBrowser.headless({driver: 'phantomjs'});
    function updateSpecs() {
        return testBundler.bundle()
            .pipe(vinylStream(jsBundleName))
            .pipe(vinylBuffer())
            .pipe(jasmineBrowser.specRunner({console: true}))
            .pipe(testServer);
    }
    testBundler.on('update', updateSpecs);
    updateSpecs();
});

Alas, this doesn't work. Right after starting the task, all tests run fine, but the next time updateSpecs is called, I get a write after end error and the task exits with status 1. This error originates from the readable-stream Node module.

As I understand it, the end event during the first run of updateSpecs leaves testServer in a state in which it doesn't accept any new inputs. Unfortunately, the Node.js streams documentation isn't very clear on how to remedy this.

I have tried breaking the pipe chain at a different place, but I got the same result, which seems to indicate this is universal behaviour for streams. I also tried stopping the end event from propagating by inserting a through-stream that didn't re-emit that event, but this prevented the tests from being run at all. Finally, I tried returning the testServer stream from the task; this stopped the error, but although the updateSpecs function gets called every time the sources change, the tests are only being run the first time the task starts. This time, the testServer simply seems to ignore the new input.

The gulp-jasmine-browser documentation suggests that the following code would work:

var watch = require('gulp-watch');

gulp.task('test', function() {
    var filesForTest = ['src/**/*.js', 'spec/**/*-test.js'];
    return gulp.src(filesForTest)
        .pipe(watch(filesForTest))
        .pipe(jasmineBrowser.specRunner())
        .pipe(jasmineBrowser.server());
});

And it goes on to suggest that you can also make this work with Browserify, but this isn't illustrated. Apparently, gulp-watch does something which causes the follow-up pipes to accept updated inputs later. How can I imitate this behaviour with watchify?

Answer 1

GitHub issue

As it turns out, it is a hard rule in Node.js that you cannot write after the end event. In addition, jasmineBrowser.specRunner(), .server() and .headless() must receive the end signal in order to actually test anything. This restriction is inherited from the official Jasmine test runner.

The example with gulp-watch from the README doesn't actually work, either, for the same reason. In order to make it work, one would have to do something similar to the working version of my watchify code in the question:

gulp.task('test', function() {
    var filesForTest = ['src/**/*.js', 'spec/**/*-test.js'];
    function runTests() {
        return gulp.src(filesForTest)
            .pipe(jasmineBrowser.specRunner())
            .pipe(jasmineBrowser.server());
    }
    watch(filesForTest).on('add change unlink', runTests);
});

(I didn't test it, but something very close to this should work.)

So whatever watching mechanism you're using, you'll always need to call .specRunner() and .server() again for every cycle. The good news is that apparently, the Jasmine server will be reused if you explicitly pass a port number:

            .pipe(jasmineBrowser.server({port: 8080}));

this also applies to .headless().