identify two symbols in a copymodule without ambiguity.

[kohlhase]

In smglom/ai/source/mod/model-based-agent.en.tex, we have

\begin{extstructure}{MBA}[model-based agent]{agent}
\begin{copymodule}{transsys}{transsys}
  \assign{actions}{\sn{agent?action}}
  \renamedecl{tm}[transition model]{tm}
  \renamedecl{states}[state]{states}
\end{copymodule}  
\symdef{sensmod}[args=2,name=sensor model]{\maincomp{S}\dobrackets{#1,#2}}
...

And this is almost what we need. Remember that an agent (from smglom/ai/source/mod/agent.en.tex)

\begin{mathstructure}{agent}
\symdef{percepts}[name=percept]{\comp{\mathcal{P}}}
\symdef{actions}[name=action]{\comp{\mathcal{A}}}
\symdef{agentfun}[args=a,name=agent function]{\maincomp{f}\dobrackets{#1}}
\notation{agent}{A}
...

has percepts, actions, and an agent function (from percept sequences to actions) to which we want to add world states and how they change wrt. to actions.
For this construction we want to identify the actions of the agent and those of the transition system.

Unfortunately, the \assign{actions}{\sn{agent?action}} above does not do that, it just maps one to the other, but we still have two symbols: agent?action and transsys?action which creates spurious ambiguities (literally hundreds in the AI notes) down the line.

I do not think that this is a feature as Dennis claims. In the context of a Model-based agent there should be only one symbol! I believe that this should be an sTeX morphisms requirement.

cc: @McEarl

[Jazzpirate]

Unfortunately, the \assign{actions}{\sn{agent?action}} above does not do that, it just maps one to the other, but we still have two symbols: agent?state and transsys?state which creates spurious ambiguities (literally hundreds in the AI notes) down the line.

Theere is no morphism with agent as a domain here, so naturally, nothing about agent is changed. If you want to delete a symbol that otherwise does exist, no morphism can do that, or should be able to do that. And that is what you're asking:
We have conceptually two different things: "The states of an agent" (in general) and "the states of an MBA viewed as a transition system". That gives us two distinct URIs. That they happen to be equal is coincidental. If their URIs are to be equal, one of them has to go and hence invalid to refer to.

I do not think that this is a feature as Dennis claims.

The "feature" is that "any symbol can be identified via a URI the reflects the path through the morphisms". You convinced me that that is more important than implicit morphisms that deliberately do not do that. I warned of the consequences of not having implicit morphisms; now we don't have implicit morphisms anymore ;)

[Jazzpirate]

Scratch that, I was about to propose a solution, but something here doesn't make sense.

the \assign{actions}{\sn{agent?action}} above does not do that

we still have two symbols: agent?state and transsys?state

...we have two states, yes, because only the actions are explicitly identified. Is there a typo in the issue or an oversight in the original tex module?

[kohlhase]

Scratch that, I was about to propose a solution, but something here doesn't make sense.
...
...we have two states, yes, because only the actions are explicitly identified. Is there a typo in the issue or an oversight in the original tex module?

This was a typo in the issue, sorry. The modules are as I want them. So propose the solution please.

BTW, I stand by my intuition that we do not want implicit morphisms. But this is also quite annoying in practice.

[Jazzpirate]

Ok, so that means the problem is that we have two actions; namely, agent?action and MBA?transsys/action, right?

Firstly: I just noticed that even having an implicit morphism wouldn't change that, actually, and for the same reason. "The action of a transition system" and "The action of an agent" are just two distinct things; that both of them make sense in an MBA means both of them have valid and distinct URIs to refer to them, and that they happen to point to the same "thing" is coincidental.

So having them have the same URI is definitely not a solution - not just because it's impossible to do, but also, because you would literally lose the ability to refer to a symbol that necessarily exists. You want to be able to refer to "the actions of the agent" in an MBA, and since an MBA is a transition system, you also want to be able to refer to "the actions of the transition system of the MBA".

The actual problem is a purely convenience problem, namely that there now are (correctly!) two symbols that are called "actions" that want (and need) to be distinguished. So if you want to make it more convenient to distinguish between them unambiguousy, the solution is to rename one of them to be sufficiently distinct. I propose an additional
\renamedecl{actions}[actions of the transition system]{} (I would hope an empty argument is valid - if not, {ts-action} or something would make sense). That should suffice to make \sn{actions} unambiguous down the line, and would still allow you to explicitly refer to "the actions of the transition system of the MBA" if you want to deliberately do so.

[kohlhase]

Ah, htis makes sense, and I like the solution. I think that I should probably try to write this up as a best practice (after testing).