Warn/error on "over ... with this"

[philrz]

tl;dr

A community user recently found the Zed language allowed them to write a lateral subquery preceded by over ... with this and the resulting overlapping uses of this values inside the subquery caused confusion. Since a lateral subquery is often used to work with the sequence of outer values via inner references to this, we recognized it would be helpful to surface a warning or error if a user's program invokes this overlapping usage, since it's probably a mistake.

Details

Repro is with Zed commit 71e35c5.

This issue was surfaced in a community Slack thread. They first started with the query shown below, which I'm executing now with our own simplified aws.json test data:

{
    "Reservations": [
        {
            "Groups": [],
            "Instances": [
                {
                    "InstanceId": "i-033bd8fe8b30bb00f",
                    "Tags": [
                        {
                            "Key": "Name",
                            "Value": "shasta"
                        }
                    ]
                }
            ]
        },
        {
            "Groups": [],
            "Instances": [
                {
                    "InstanceId": "i-019bcf4e640743aa6",
                    "Tags": [
                        {
                            "Key": "Name",
                            "Value": "phil-win2019"
                        }
                    ]
                }
            ]
        }
    ]
}

$ zq -version
Version: v1.17.0-20-g71e35c5d

$ zq -z 'over Reservations | over Instances
  | over Tags with obj=this => ( where Key=="Name" | {name: this.Value, id:obj.InstanceId} )' aws.json
{name:"shasta",id:"i-033bd8fe8b30bb00f"}
{name:"phil-win2019",id:"i-019bcf4e640743aa6"}

The user then remarked:

Huh - I noticed with a couple small edits - I could say with this (vs. with obj=this) in the lateral subquery ... and by removing the this from this.Value since those collide now ... it still does work:

zq -z '
 over Reservations | over Instances
 | over Tags with this => ( where Key=="Name" | {name: Value, id:this.InstanceId} )' aws.json
{name:"shasta",id:"i-033bd8fe8b30bb00f"}
{name:"phil-win2019",id:"i-019bcf4e640743aa6"}

But - that feels ... dangerous? Or odd or something? Maybe it's not.

The way he was making an implied field reference to just Value was allowing him to still access the "inner" data that would normally be referenced as this.Value. He ultimately figured out on his own what was going on.

I guess it just means that this is going to stomp the "implied" this inside the subquery. that's at least what it does:

zq -z '
 over Reservations | over Instances
 | over Tags with this => ( where Key=="Name" | {name: this.Value, id:this.InstanceId} )' aws.json
{name:error("missing"),id:"i-033bd8fe8b30bb00f"}
{name:error("missing"),id:"i-019bcf4e640743aa6"}

However, since it required a close reading of the docs to clear up the confusion, some messaging from the tooling would have also been helpful. In the user's opinion:

off the cuff, i’d vote for warning rather than error … it’s not broken per se, just … stompy :hiking_boot: :smile: