print_backdate_admonition may need a backtrace

[timholy]

The binding partitioning work introduced stricter rules for accessing globals, including a scary warning from

julia/src/module.c

Lines 775 to 784 in 13311f3

	static NOINLINE void print_backdate_admonition(jl_binding_t *b) JL_NOTSAFEPOINT
	{
	jl_safe_printf(
	"WARNING: Detected access to binding `%s.%s` in a world prior to its definition world.\n"
	" Julia 1.12 has introduced more strict world age semantics for global bindings.\n"
	" !!! This code may malfunction under Revise.\n"
	" !!! This code will error in future versions of Julia.\n"
	"Hint: Add an appropriate `invokelatest` around the access to this binding.\n",
	jl_symbol_name(b->globalref->mod->name), jl_symbol_name(b->globalref->name));
	}

My experience is that it's been quite difficult to pinpoint the problematic line without a backtrace. Another option would be to have an option to turn the warning into an error.

[vtjnash]

How strict do we want to be? We could do this:

diff --git a/src/module.c b/src/module.c
index 7d77e933f81..a54384967aa 100644
--- a/src/module.c
+++ b/src/module.c
@@ -773,6 +773,8 @@ JL_DLLEXPORT jl_module_t *jl_get_module_of_binding(jl_module_t *m, jl_sym_t *var
 
 static NOINLINE void print_backdate_admonition(jl_binding_t *b) JL_NOTSAFEPOINT
 {
+    if (jl_options.depwarn == JL_OPTIONS_DEPWARN_ERROR)
+        jl_undefined_var_error(b->globalref->name, (jl_value_t*)b->globalref->mod);
     jl_safe_printf(
         "WARNING: Detected access to binding `%s.%s` in a world prior to its definition world.\n"
         "  Julia 1.12 has introduced more strict world age semantics for global bindings.\n"
diff --git a/base/errorshow.jl b/base/errorshow.jl
index c53c605cbfb..b1b7a083eb0 100644
--- a/base/errorshow.jl
+++ b/base/errorshow.jl
@@ -1163,6 +1163,8 @@ function UndefVarError_hint(io::IO, ex::UndefVarError)
                 print(io, "\nSuggestion: check for spelling errors or missing imports.")
             elseif Base.is_some_explicit_imported(kind)
                 print(io, "\nSuggestion: this global was defined as `$(Base.partition_restriction(bpart).globalref)` but not assigned a value.")
+            elseif kind === Base.PARTITION_KIND_BACKDATED_CONST
+                print(io, "\nSuggestion: this const was attempted to be used before it was defined (new Julia v1.12 rule).")
             end
         elseif scope === :static_parameter
             print(io, "\nSuggestion: run Test.detect_unbound_args to detect method arguments that do not fully constrain a type parameter.")

resulting in:

$ ./julia --depwarn=error -E 't = @task x; const x = 4; fetch(schedule(t))'
ERROR: TaskFailedException
Stacktrace:
 [1] #wait#581
   @ ./task.jl:363 [inlined]
 [2] wait
   @ ./task.jl:360 [inlined]
 [3] fetch(t::Task)
   @ Base ./task.jl:525
 [4] top-level scope
   @ none:1
 [5] eval(m::Module, e::Any)
   @ Core ./boot.jl:489
 [6] exec_options(opts::Base.JLOptions)
   @ Base ./client.jl:296
 [7] _start()
   @ Base ./client.jl:560

    nested task error: UndefVarError: `x` not defined in `Main`
    Suggestion: this const was attempted to be used before it was defined (new Julia v1.12 rule).
    Stacktrace:
     [1] (::var"#5#6")()
       @ Main ./none:1

[timholy]

I like it. The only minor weakness I see with this proposal is that you can't get lineinfo unless you make it an error, so it's not possible to use it like @depwarn which (nominally) gives line info even when it's just a warning. That said, I'd say the convenience is minor, and (at least as of Julia 1.10) the depwarn line number printing is still sufficiently hit-or-miss that I often turn them into errors anyway.

Possible alternate phrasing: "Suggestion: define the constant before running code that uses it (stricter Julia v1.12+ rule)."

[vtjnash]

It looks like the hint code is also not currently supporting binding ages like method printing does (to alert you if there is a binding in a different world with different values) which is likely a separate issue with the resulting hint text

[vtjnash]

I marked for triage/needs decision, since I want folks to be on-board with making this controlled by the existing depwarn error flag before we commit to doing that

[Keno]

I do think the UndefVarError one is reasonable, because that's closest to the behavior you'll get after we remove the deprecation.