Cannot pass dynamic values to initializers

[Chuck]

The following code gives false compilation errors on the commented line:

class X

	def init
		.init(0, 'aoeu')
		
	def init(i as int, s as String)
		pass

	def foo(i as int, s as String)
		pass

	def main is shared
		a = 1 to dynamic
		b = 'aoeu' to dynamic
		x = X(a, b)  # test passing dynamic to an initializer
		x = X()
		x.foo(a, b)

[eric.sellon]

Patch file attached. Hope it helps.

[Chuck]

With this patch, PostCallExpr? now has a _computeBestOverload which is not quite the same as CallExpr?. Are the differences intentional or accidental? Do you think they could share one _computeBestOverload in a common base class? I ask because that's originally what I envisioned for this fix.

Otherwise, it seems we have two method overload resolution algorithms to maintain.

Let me know your thoughts.

[Chuck]

ping.

Would love to close this.

[eric.sellon]

Sorry. I've been busy lately and just today got time again to work on this. I figured a few hours would allow me to resolve this today, but I was wrong. So, I'm still working on getting all the tests passed with a single computeBestOverload function.

[Chuck]

Thanks for the update.

[Chuck]

Oh and if you want some IRC conversation, you have stick around for more than 2 minutes. Sometimes we're AFK for a period of time. :-)

[eric.sellon]

If you use the attached debug help patch file and then try to compile 100-attributes.cobra, you'll see that symbolForName is returning something different on the second FallThroughException?. Please advise. Thank you.

[Chuck]

I'm looking tonight.

[Chuck]

I don't know if I have exactly the answer you were looking for, but I definitely have the info you need:

• First, I noticed that the Param printed with didBindInt=false which is bad news. We should not be examining the interface of something (such as its .type) if it never went through the "bind interface" phase.
• So I added require .bindInt to the MemberOverload.computeBestOverload method. I ran the compiler and this failed as I was expecting. Why?
• In Expr.cobra there is the line possibleCalls = theClass.symbolForName('cue.init') which is the source of evil. Change it to .memberForName and everything is good.
• .memberForName is more advanced than .symbolForName. It "preps" library types if they are unprepared due to not having done a .bindInh and .bindInt. It also includes extension methods. .memberForName is typically what you want.
• I realize this is not obvious. Sorry for any confusion.
• Also in the future, you may prefer trace x over print x for debugging.

[eric.sellon]

Great. Thanks for looking at it. I hope to try that out tonight.

[eric.sellon]

Updated patch file. Should be good.

[Chuck]

Thanks. Applied in changeset:2167

One big item of feedback: in tests like 310-dynamic-init-params.cobra, use assert instead of print to validate the behavior. This way the testing is automatic.

See changeset:2167 for details on how I changed this.