Don't include lhs of := in results of predict()

[penelopeysm]

Closes #765

Specifically, lhs := rhs expressions are treated here:

DynamicPPL.jl/src/compiler.jl

Lines 379 to 389 in 6657441

	# Modify the assignment operators.
	args_assign = getargs_coloneq(expr)
	if args_assign !== nothing
	L, R = args_assign
	return Base.remove_linenums!(
	generate_assign(
	generate_mainbody!(mod, found, L, warn),
	generate_mainbody!(mod, found, R, warn),
	),
	)
	end

This in turn checks for is_extracting_values(context):

DynamicPPL.jl/src/compiler.jl

Lines 394 to 404 in 6657441

	function generate_assign(left, right)
	right_expr = :($(TrackedValue)($right))
	tilde_expr = generate_tilde(left, right_expr)
	return quote
	if $(is_extracting_values)(__context__)
	$tilde_expr
	else
	$left = $right
	end
	end
	end

Currently, this function only returns true for ValuesAsInModelContext, and returns false for every other context.

DynamicPPL.jl/src/values_as_in_model.jl

Lines 40 to 45 in 6657441

	is_extracting_values(context::ValuesAsInModelContext) = true
	function is_extracting_values(context::AbstractContext)
	return is_extracting_values(NodeTrait(context), context)
	end
	is_extracting_values(::IsParent, ::AbstractContext) = false
	is_extracting_values(::IsLeaf, ::AbstractContext) = false

This PR adds a new boolean field to ValuesAsInModelContext which controls whether is_extracting_values(context) is true. For predict, this field is set to false, so that lhs := rhs lines are not included in the resulting chain.

Instead of providing a default value, I've opted to force callers of values_as_in_model specify whether they want the := variables, so this is a minor release.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.12%. Comparing base (b7fd9ea) to head (a055574).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #766      +/-   ##
==========================================
+ Coverage   86.05%   86.12%   +0.06%     
==========================================
  Files          36       36              
  Lines        4325     4324       -1     
==========================================
+ Hits         3722     3724       +2     
+ Misses        603      600       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[coveralls]

Pull Request Test Coverage Report for Build 12598010488

Details

• 7 of 7 (100.0%) changed or added relevant lines in 2 files are covered.
• 51 unchanged lines in 11 files lost coverage.
• Overall coverage increased (+0.07%) to 86.204%

---

Files with Coverage Reduction	New Missed Lines	%
src/varnamedvector.jl	1	88.25%
src/sampler.jl	1	94.55%
src/utils.jl	2	73.2%
src/contexts.jl	3	27.27%
src/values_as_in_model.jl	3	70.37%
src/distribution_wrappers.jl	4	41.67%
src/model.jl	5	77.39%
src/varinfo.jl	6	84.62%
src/simple_varinfo.jl	6	81.96%
src/compiler.jl	8	86.58%

Totals
Change from base Build 12597566459:	0.07%
Covered Lines:	3724
Relevant Lines:	4320

---

💛 - Coveralls

[sunxd3]

very sensible, thanks!

[torfjelde]

Didn't see this as I was on vacation, but if this had been done as a kwarg with default to true, this could have been a non-breaking change?