[BUG] regex group variables reset on subroutine transitions

[richardmarshall]

Kind of proposals

• [x] Simulator
• [x] Testing

Describe the problem

When changing scope to a new subroutine via call or a function expression the interpreter resets the re.group.# variables in the same way it does for local variables. This is not correctly following the behavior of Fastly.

Additionally accessing a re group variable before performing a regex match currently produces an undefined variable interpreter error instead of returning a NotSet value.

VCL code that cause the problem / reproduceable

sub foo {
  set req.http.foo1 = re.group.1;
  if (req.http.test ~ "([^.]+)") {}
  set req.http.foo2 = re.group.1;
}

sub bar {
  set req.http.bar1 = re.group.1;
  if (req.http.test ~ "(.*)") {}
  set req.http.bar2 = re.group.1;
  call foo;
  set req.http.bar3 = re.group.1;
}

  ●  [RECV] test

    Assertion Error: Assertion error: expect=aoeu.aoeu, actual=
    Actual Value: 

    5|   assert.equal(req.http.bar2, "aoeu.aoeu");
    6|   assert.equal(req.http.foo1, "aoeu.aoeu");
    7|   assert.equal(req.http.foo2, "aoeu");

Expected behavior

https://fiddle.fastly.dev/fiddle/a3a1975b

[ysugimoto]

Oh, I didin't know this spec...

I think we should follow the behavior in the interpreter but the linter should report an error as WARNING because this spec causes a potential bug.

[richardmarshall]

@ysugimoto A linter warning is a good idea. I updated #294 to include a fix for the interpreter since I was already moving around the subroutine state changes. Can make another PR that adds warning logic to the linter.