Stacktraces reported by Minetest blame the wrong mod

[Lars Müller]

Minetest version

5.5.0 and earlier; presumably 5.6-dev as well

Summary

See minetest-mods/wielded_light#21: The stacktrace clearly shows mesecons is responsible, but the error message blames the wielded_light mod.

Steps to reproduce

//lua table.insert(minetest.registered_globalsteps, function() _ = nil + nil end)

[rubenwardy]

"From mod X" is saying the get_current_modname value - wielded_light is the mod that's currently loading. It's not based on stacktrace

It should probably say "error whilst loading mod" rather than "error from mod"

[Lars Müller]

"From mod X" is saying the get_current_modname value - wielded_light is the mod that's currently loading. It's not based on stacktrace

It should probably say "error whilst loading mod" rather than "error from mod"

This should be after load time though, unless I misinterpret "in callback environment_Step()"?

On the other hand, the stacktrace clearly points to a chunk that would most likely be loaded at load time (although it could be reloaded at runtime); if this were at load time though, it should correctly detect mesecons as responsible as the mesecons initialization isn't loaded by the wielded light mod.

[sfan5]

core.run_callbacks sets the correct mod origin on a best effort basis:

minetest/builtin/game/register.lua

Lines 421 to 425 in 480d5f2

 local origin = core.callback_origins[callbacks[i]] 
 if origin then 
 	core.set_last_run_mod(origin.mod) 
 end 
 local cb_ret = callbacks[i](...) 

As you can see if a callback doesn't have a mod origin set the last one will stick around. But without a concrete reproducer there is nothing that can be done about this issue, since it generally works.

[Lars Müller]

core.run_callbacks sets the correct mod origin on a best effort basis:

minetest/builtin/game/register.lua

Lines 421 to 425 in 480d5f2

 local origin = core.callback_origins[callbacks[i]] 
 if origin then 
 	core.set_last_run_mod(origin.mod) 
 end 
 local cb_ret = callbacks[i](...) 

As you can see if a callback doesn't have a mod origin set the last one will stick around. But without a concrete reproducer there is nothing that can be done about this issue, since it generally works.

That looks like an obvious bug to me. It should clear the "origin" using core.set_last_run_mod"??" or the like.

[Lars Müller]

A simple way to reproduce this:

//lua table.insert(minetest.registered_globalsteps, function() _ = nil + nil end)