v1.12: JuliaInterpreter optimization seems to badly interact with new compressed linetable format

[aviatesk]

It looks like this "optimization" introduces a very bad interaction with the new lineinfo format:

JuliaInterpreter.jl/src/optimize.jl

Lines 76 to 92 in 28dd37c

	# TODO: because of builtins.jl, for CodeInfos like
	# %1 = Core.apply_type
	# %2 = (%1)(args...)
	# it would be best to *not* resolve the GlobalRef at %1
	## Replace GlobalRefs with QuoteNodes
	for (i, stmt) in enumerate(code.code)
	if isa(stmt, GlobalRef)
	code.code[i] = lookup_global_ref(stmt)
	elseif isa(stmt, Expr)
	if stmt.head === :call && stmt.args[1] === :cglobal # cglobal requires literals
	continue
	else
	lookup_global_refs!(stmt)
	code.code[i] = lookup_getproperties(code.code, stmt)
	end
	end
	end

julia> using JuliaInterpreter

julia> fnone1() = nothing;

julia> let m = only(methods(fnone1))
           JuliaInterpreter.FrameCode(m, JuliaInterpreter.get_source(m); optimize=true).src.debuginfo
       end
Core.DebugInfo(Symbol("REPL[2]"), nothing, svec(), "\x01\0\0\0\0\0\0\0\0\0\0\0\x01\x01\0")

julia> fnone2() = nothing;

julia> let m = only(methods(fnone2))
           JuliaInterpreter.FrameCode(m, JuliaInterpreter.get_source(m); optimize=true).src.debuginfo
       end
Core.DebugInfo(Symbol("REPL[4]"), nothing, svec(), "")

Originally posted by @aviatesk in #606 (comment)

[timholy]

Is this still relevant? I get the same output in both cases. And it doesn't depend on the value of optimize.

[simeonschaub]

A better reproducer:

julia> using JuliaInterpreter

julia> fnone1() = nothing;

julia> let m = only(methods(fnone1))
           JuliaInterpreter.FrameCode(m, JuliaInterpreter.get_source(m); optimize=true).src.debuginfo
       end
Core.DebugInfo(Symbol("REPL[2]"), nothing, svec(), "\x01\0\0\0\0\0\0\0\0\0\0\0\x01\x01\0")

julia> @interpret (() -> nothing)()

julia> fnone2() = nothing;

julia> let m = only(methods(fnone2))
           JuliaInterpreter.FrameCode(m, JuliaInterpreter.get_source(m); optimize=true).src.debuginfo
       end
Core.DebugInfo(Symbol("REPL[5]"), nothing, svec(), "")

Note that if the nothing @interpret (() -> nothing)() is replaced with anything else (e.g. @interpret (() -> 7)()) this behavior disappears.

[simeonschaub]

It also doesn't seem to have anything to do with those particular optimizations since it still happens with optimize=false:

julia> using JuliaInterpreter

julia> fnone1() = nothing;

julia> let m = only(methods(fnone1))
           global fc = JuliaInterpreter.FrameCode(m, JuliaInterpreter.get_source(m); optimize=false)
           fc.src.debuginfo
       end
Core.DebugInfo(Symbol("REPL[2]"), nothing, svec(), "\x01\0\0\0\0\0\0\0\0\0\0\0\x01\x01\0")

julia> let fr = JuliaInterpreter.prepare_frame(fc, [], Core.svec())
           JuliaInterpreter.finish_and_return!(fr)
       end

julia> fnone2() = nothing;

julia> let m = only(methods(fnone2))
           JuliaInterpreter.FrameCode(m, JuliaInterpreter.get_source(m); optimize=false).src.debuginfo
       end
Core.DebugInfo(Symbol("REPL[5]"), nothing, svec(), "")

[simeonschaub]

Reported upstream as JuliaLang/julia#55688 with a non-JuliaInterpreter reproducer. Seems like we can close this issue here