Reasons we need to use "empty" or null values. about arm-ttk

[bmoore-msft]

I think subnets[] may be something we need to except, but in this case I think using json('null') is correct?

[marrobi]

Not sure subnets[] would work. As it can be different resources that take an array. Firewall rules for example.

Are you saying json('null') should work? Sure I tried it, but can try again might have been the union function that didn't like it.

Can give it another go tomorrow.

[bmoore-msft]

oic - it's flagging the empty variable...

Instead of an empty variable use: json('[]') in your "else" side of the if() statement

[marrobi]

So json('[]'):

"subnets": "[union(variables('defaultSubnets'),if(parameters('deployVnetGateway'), variables('vnetGatewaySubnet'),json('[]') ),if(parameters('deployBastion'), variables('bastionSubnet'),json('[]') ))]",

Fails ARM-TTK due to empty value.

and json('null'):

"subnets": "[union(variables('defaultSubnets'),if(parameters('deployVnetGateway'), variables('vnetGatewaySubnet'),json('null') ),if(parameters('deployBastion'), variables('bastionSubnet'),json('null') ))]",

Gives:

Error: Code=InvalidTemplate; Message=Deployment template validation failed: 'The template variable 'subnets' is not valid: Template language function 'union' expects parameters of
     | the same type, but found multiple types..

[amergey]

Very similar needs for us
We need to "transform" a variable list of parameters into an array so we use this kind of thing (which is simplified to illustrate here) :
[union(array('val'),if(empty(parameters('param')),json('[]'),array(parameters('anotherVal'))))]

this is failing because json('[]') is considered as an empty property for "Template Should Not Contain Blanks"

There is probably alternative way to do achieve that without json('[]'), at a price of a much more complex logic to follow

[bmoore-msft]

Ack... json('[]') was working for me (i.e. tests didn't flag it) due to user error (I used it in an output which is already excepted). Anyway....

So I could think of a few options here:

1. allow certain variable names to be empty
   Generally I think if you use an empty Var you know why you're doing it and it's intentional... the flip side is that you would miss come cases where you don't want it directly assigned to a property. So I could say any variable that is named something like "empty*" would be ignored by the test. That requires you to know about the "magic" naming convention to bypass the test

2. except the json() function specifically
   Indirectly we do this for json('null') today, we could do it (easily) for any time the empty json is not preceded by colon. This would required the template author to use the json() function anytime you want to use empty... it would also mean the test would not flag incorrect use of empty, but that's extremely rare in practice.

I think I like #2 - thoughts?

[bmoore-msft]

or both?

[amergey]

I am not the initial reporter of the issue, but I also like #2 :)

[marrobi]

2 works for me. Unless there are instances other than empty arrays that can be dealt with using the same logic, such as '{}'.

[bmoore-msft]

yes, I think the logic would work the same for empty objects, strings, nulls... So the cases we would catch are where you literally specify an empty property (regardless of type)... which TBH is the common case and the common mistake. If you've authored it as an expression, you're intentionally doing so.

[bmoore-msft]

added an approach in #62