PhantomJs: ??? in onError + JSCom has been closed #1555
Comments
japgolly
changed the title
|
Thanks for the report. I'll probably not be able to look into it before 26th, though. |
|
np :) I'll lock in a commit in the reproduction instructions above and switch off phantomjs for now. |
|
Sweet!! Finally we can reproduce this. |
|
I can't reproduce this :( It seems this is really just a race somewhere... |
|
Oh no really? You definately checked it out to aed8b40 (cos I switched to Node JS after that commit). For me it happens 100% of the time. Maybe there's an environmental factor... |
That might be it... :S |
|
I've just started getting this in a completely different (private) repo. :( |
|
This has something to do with object initialisation order I believe. |
|
Is there anyway we can prioritise this? This issue follows me around from project to project. |
|
It always mentions |
|
This makes me want to swear. A lot. PhantomJS releases are very few and far between... Could we pretty-please have a workaround in SJS for this? Like a configuration setting that determines which |
|
If you can come up with a work-around in doWriteLine that still uses In any case I'll schedule a least a work-around for 0.6.3. |
|
I wonder what we could do. Thinking slightly bigger, it might be handy if developers were able to control the behaviour of What do you think about |
|
You can mutate |
|
How? |
|
Ah, right. Well I guess we need to add those. :-s |
|
Ah, I thought I was missing something again. :) Raised #1598 So the above will facilitate users being able to workaround it themselves (everywhere in source). What should we do to prevent such a need? Maybe redirect |
|
You can subclass |
|
Indeed. Actually, I think we should rather solve this particular problem on a JS level (so the Scala code remains as VM agnostic as possibled). Stderr will automatically be redirected to class PhantomJS2Env(
phantomjsPath: String = "phantomjs",
addArgs: Seq[String] = Seq.empty,
addEnv: Map[String, String] = Map.empty,
val autoExit: Boolean = true,
jettyClassLoader: ClassLoader = null
) extends PhantomJSEnv(phantomjsPath, addArgs, addEnv, autoExit, jettyClassLoader) {
override protected def vmName: String = "PhantomJS2"
private val consoleNuker = new MemVirtualJSFile("consoleNuker.js")
.withContent("console.error = undefined;")
override def jsRunner(classpath: CompleteClasspath, code: VirtualJSFile,
logger: Logger, console: JSConsole): JSRunner = {
new PhantomRunner(classpath, code, logger, console) {
override protected def initFiles(): Seq[VirtualJSFile] = super.initFiles :+ consoleNuker
}
}
override def asyncRunner(classpath: CompleteClasspath, code: VirtualJSFile,
logger: Logger, console: JSConsole): AsyncJSRunner = {
new AsyncPhantomRunner(classpath, code, logger, console) {
override protected def initFiles(): Seq[VirtualJSFile] = super.initFiles :+ consoleNuker
}
}
override def comRunner(classpath: CompleteClasspath, code: VirtualJSFile,
logger: Logger, console: JSConsole): ComJSRunner = {
new ComPhantomRunner(classpath, code, logger, console) {
override protected def initFiles(): Seq[VirtualJSFile] = super.initFiles :+ consoleNuker
}
}
}And then use this, instead of the standard PhantomJS: // Set other stuff as you require
postLinkJSEnv := new PhantomJS2Env(jettyClassLoader = scalaJSPhantomJSClassLoader.value) |
|
That looks like it'd do the job. Is this something you envision being part of Scala.JS? It's quite a large snippet :) Also that's a lot of boilerplate just to add some initial-JS. It might be worthwhile to add an |
|
I don't know if this should be part of Scala.js. Are you working with PhantomJS 2.0 or with a pre-release version? Maybe we should upgrade to PhantomJS 2.0 anyways internally... @sjrd what do you think? That would also give us proper websockets, meaning we could upgrade to Jetty9. |
|
Upgrading to PhantomJS 2.0 would be nice, but isn't it a bit too early to drop support for PhantomJS 1.x? |
|
It's PhantomJS 2.0 final. The issue is marked as a regression so it must be a problem in certain version of PhantomJS 1.x as well. |
|
Is the error we get from PhantomJS catchable? If yes, we can easily monkey patch
In any case, this will be a testing nightmare for us, since we'll have to test with multiple versions of PhantomJS :( |
|
What do we want to do about this? It's definitely a crasher in PhantomJS, so it must eventually be fixed. Maybe we could add an overrideable method class PhantomJS2Env(
phantomjsPath: String = "phantomjs",
addArgs: Seq[String] = Seq.empty,
addEnv: Map[String, String] = Map.empty,
val autoExit: Boolean = true,
jettyClassLoader: ClassLoader = null
) extends PhantomJSEnv(phantomjsPath, addArgs, addEnv, autoExit, jettyClassLoader) {
override protected def vmName: String = "PhantomJS2"
private val consoleNuker = new MemVirtualJSFile("consoleNuker.js")
.withContent("console.error = undefined;")
override protected def customInitFiles(): Seq[VirtualJSFile] =
super.customInitFiles() :+ consoleNuker
}@japgolly Would that be acceptable until they fix the issue in PhantomJS? Should we provide |
I don't think we should. In the long run, we should provide proper support for PhantomJS 2.0, with RFC 6455 websockets and Jetty9. So we might will have to envision to create a new separate environment. In the meantime, adding |
This will simplify the user-space workaround for scala-js#1555. It can also be generally useful.
This will simplify the user-space workaround for scala-js#1555. It can also be generally useful.
|
I would leave this one open. After all it's not solved. We can cross-reference this ticket from the more general PhantomJS 2.0 one. |
|
OK |
Anything that allows me to workaround this is acceptable to me, but that's my selfish view :) It might be better to have something provided because anyone who uses PhantomJS 2 is going to have this problem, and it could be a very long time before it's fixed upstream. |
Upgrading to Scala.JS 0.6.3 to workaround scala-js/scala-js#1555 seems to fix it.
|
Using the Even with this fix applied however, this issue still occurs in rare occasions (like when there is an error initialising an object in a test). #1702 will address this. If we apply #1702 and make |
Was hitting some issues (with uncommitted code) with PhantomJS. zenoamaro/react-quill#87 was making React run console.error, which blows up PhantomJS. scala-js/scala-js#1555 ariya/phantomjs#13112 Running tests in Firefox seems more stable, though slower. Unfortunately, I can't run headless firefox on Mac, which is annoying. There is no X11 version of firefox available via brew (Homebrew/legacy-homebrew#9150), or anywhere else apparently (https://trac.macports.org/ticket/42087)
gzm0
added
the
wontfix
|
Closing this as wontfix, since we'll be dropping PhantomJS. |
This will simplify the user-space workaround for scala-js/scala-js#1555. It can also be generally useful.
Reproducies:
It's been failing ever since I switched
jsEnvto phantom js.I get (abbreviated):
When I run
last core-js/test:testI get:The text was updated successfully, but these errors were encountered: