make sure to use only the RegExpProxy interface

[rbri]

No description provided.

[p-bakker]

Am I correct in assuming that args[0] can never be something considered a regex if ScriptRuntime.getRegExpProxy(cx) returns null? Otherwise wouldn't these changes modify the behavior slightly?

[rbri]

Hi @p-bakker,

there are already some places in the code that checking that the regexpproxy is not null - without having one the whole regex support is not working and crashing. See org.htmlunit.corejs.javascript.ScriptRuntime.checkRegExpProxy(Context) and the caller of this method.

org.htmlunit.corejs.javascript.ScriptRuntime.setRegExpProxy(Context, RegExpProxy) - the only way to set the proxy does also an null check and throws if not null.

The only way to get null here is to remove the default from the classpath/classolader - see the lazy initialization in org.htmlunit.corejs.javascript.Context.getRegExpProxy()

I think we can assume here that the proxy is configured...

Hope that is enough text to get this merged

[p-bakker]

Could you add some description on this PR to explain the reasoning behind it?

[rbri]

Could you add some description on this PR to explain the reasoning behind it?

At least i can try...

If you like to customize (as an project that uses Rhino) the current regex implementation you have to replace to things - the RegExpProxy implementation and the NativeRegExp impl. Both replacements can be done by subclassing the existing rhino classes or by creating your own. HtmlUnit does exactly by following the subclassing path.

Now i have started to switch to a more advanced replacement, and soon found out, that it might be more reasonable to replace both implementations. The api makes that possible - but we still have this instanceof check in our source code. Therefore this PR fixes this by using the reProxy.isRegExp() instead (and using this method in this scenario is exactly what the method is for.

Ok, and there is another (major from my point of view) improvement in this PR - i have added ArchUnit to our test suite as starting point to be able to enforce some design rules for the whole code base. I use ArchUnit since some years for HtmlUnit and it is a great help....

[p-bakker]

Will respond more indepth next week (sorta off this week), but given your understanding of the proxy pattern used here can you comment on the seemingly improved performance after implementing the same pattern for quasi call sites, as suggested here

[gbrail]

I will look at this too. In the meantime... I'm interested in what you did with HtmlUnit. My basic and very naive question is this -- does Rhino still need its own custom regexp implementation or could we use the java.util.regexp package instead?

[gbrail]

Also, the design rule check is interesting and it'd be interesting to see if we should use it in other places.

However, we have modules now. The org.mozilla.javascript.regexp package is currently not exported, so no one outside the "rhino" project can depend on it now. And if we wanted to be even more pedantic, we could consider moving this package to a separate module, so that we can guarantee that the "rhino" module builds without it. That would complicate people's builds, though, so it's debatable whether it's a good idea.

[p-bakker]

or could we use the java.util.regexp package instead?

To my knowledge they never were 100% compatible and on the other side, a lot of new features have recently been added to Regex in EcmaScript and more enhancements are on the way. So when looking into this we need to also take into account new Regex features being added to Regex in EcmaScript that Java may not support

Having said that, would be great if we could replace the engine to gain performance and be more ES-compatible!

[rbri]

I'm interested in what you did with HtmlUnit. My basic and very naive question is this -- does Rhino still need its own custom regexp implementation or could we use the java.util.regexp package instead?

As far as i rember there are some fundamental differences between js regexp and java regexp - js has supports more things and therefore there is no chance to map them all to java.

if you like you can start here https://github.com/HtmlUnit/htmlunit/tree/master/src/main/java/org/htmlunit/javascript/regexp, but for various reasons this is full of hacks

[rbri]

However, we have modules now. The org.mozilla.javascript.regexp package is currently not exported, so no one outside the "rhino" project can depend on it now. And if we wanted to be even more pedantic, we could consider moving this package to a separate module, so that we can guarantee that the "rhino" module builds without it. That would complicate people's builds, though, so it's debatable whether it's a good idea.

The idea is not so much about hoping others replace the regexp stuff in there impl, my idea was more to be able to offer a more mordern and complete impl for ES6 and use the existing one for the old stuff and backward compatibility.

[gbrail]

I am sorry that this one ended up at the bottom of my queue, and I don't know why not. I was able to merge it and resolve the conflicts, but I see some failing tests, so I'm going to ask for help to get it rebased on the latest or something close to it. Any chance that you have time for that?

[rbri]

Will have a look

[rbri]

@gbrail a new pr is on the way not sure why this one has passed all the test but i have a new one that again is hopefully green

[rbri]

PR #1580 is ready

[gbrail]

Thanks for the new PR!

[rbri]

@gbrail thanks for taking care of this - will try to make this alternative regexp impl after my vacation