printall_recurse - do not descend into untagged unions

docs/changelog.txt

@@ -67,6 +67,7 @@ Template for new versions:
 ## API
 
 ## Lua
+- ``printall_recurse``: no longer recurses into untagged unions, which have no way to identify their live member; fixes crashes when printing structures that contain them, such as ``df.army_controller``
 
 ## Removed
 

docs/dev/Lua API.rst

@@ -3960,6 +3960,9 @@ environment by the mandatory init file dfhack.lua:
 * ``printall_recurse(obj)``
 
   If the argument is a lua table or DF object reference, prints all fields recursively.
+  The fields of untagged unions are not printed since only one member of such a
+  union is live at a time and there is no way to identify which one, so
+  accessing any other member is undefined behavior.
 
 * ``copyall(obj)``
 

library/lua/dfhack.lua

@@ -367,6 +367,26 @@ local function print_fields(value, seen, indent, prefix)
     return 0
 end
 
+-- Recursing into the members of an untagged union is unsafe: only one member
+-- is live at a time and there is no tag to identify which one, so accessing
+-- any of the others is undefined behavior and can crash the game. Tagged
+-- unions are safe to iterate since their iterators only produce the live
+-- member.
+local function is_untagged_union(value)
+    if df.isvalid(value) ~= 'ref' then
+        return false
+    end
+    local vtype = value._type
+    if type(vtype) ~= 'table' or not vtype._union then
+        return false
+    end
+    -- a tagged union's iterator produces at most the one live member; an
+    -- untagged union's iterator produces all of them
+    local count = 0
+    safe_iterate(value, pairs, function() count = count + 1 end)
+    return count > 1
+end
+
 -- This should be same as print_array but userdata doesn't compare equal even if
 -- they hold same pointer.
 local function print_userdata(printfn, value, seen, indent)
@@ -375,10 +395,15 @@ local function print_userdata(printfn, value, seen, indent)
     dfhack.println(strvalue)
     if seen[strvalue] then
         dfhack.print(prefix)
-        dfhack.println('<Cyclic reference! Skipping fields>\n')
+        dfhack.println('<Cyclic reference! Skipping fields>')
         return 0
     end
     seen[strvalue] = true
+    if is_untagged_union(value) then
+        dfhack.print(prefix)
+        dfhack.println('<Untagged union! Skipping fields>')
+        return 0
+    end
     return print_fields(value, seen, indent, prefix)
 end
 
@@ -387,7 +412,7 @@ local function print_array(printfn, value, seen, indent)
     dfhack.println(tostring(value))
     if seen[value] then
         dfhack.print(prefix)
-        dfhack.println('<Cyclic reference! skipping fields>\n')
+        dfhack.println('<Cyclic reference! skipping fields>')
         return 0
     end
     seen[value] = true

test/library/print.lua

@@ -190,6 +190,31 @@ function test.printall_recurse()
     end)
 end
 
+function test.printall_recurse_untagged_union()
+    local t = df.army_controller:new()
+    dfhack.with_temp_object(t, function()
+        printall_recurse(t.data)
+        expect.eq(3, mock_print.call_count)
+        validate_patterns(1,
+            {'^<army_controller.T_data: ',
+             '', '^<Untagged union'})
+    end)
+end
+
+function test.printall_recurse_tagged_union()
+    local EQ = '^%s+= $'
+    local t = df.unit_patient_profile_completed_jobst:new()
+    dfhack.with_temp_object(t, function()
+        -- RecoverWounded maps to the 'bed_id' member via tag_op_history
+        t.job_type = df.job_type.RecoverWounded
+        printall_recurse(t.info)
+        expect.eq(5, mock_print.call_count)
+        validate_patterns(1,
+            {'^<unit_patient_profile_completed_jobst.T_info: ',
+             '', '^bed_id$', EQ, '^%-1$'})
+    end)
+end
+
 function test.printall_recurse_cyclic_userdata()
     local t = df.job_list_link:new()
     dfhack.with_temp_object(t, function()