Format: accept a named pattern or a blank as the argument, e.g. Format[t_h] := ...

[tombouctou]

eval_assign_format replaced the left-hand side of a Format definition by its head before looking for the tag. For a named pattern or a blank the head is Pattern or Blank, so the assignment failed:

In[1]:= Format[A_G] := "g" /; Length[A] > 1
SetDelayed::write: Tag Pattern in A_G is Protected.
Out[1]= $Failed

In WMA this defines a format for G; Format[_G] and Format[t:G[__]] work the same way.

This PR keeps named patterns and blanks as they are, so the tag (G) is found inside them by collect_lhs_names_and_upset_dont_allow_custom, as for other assignments. Related: #898.

Tests: new cases in test/builtin/assignments/test_assignment.py (test_assignment).

🤖 Generated with Claude Code

[mmatera]

LGTM

[rocky]

@mmatera I am guessing then you have no problem with accepting Claude Code commits, right?

[mmatera]

@mmatera I am guessing then you have no problem with accepting Claude Code commits, right?

As I just said in #1991 (comment), in principle I am OK with it, as far as it does not flood our review capacity.

[rocky]

@mmatera I am guessing then you have no problem with accepting Claude Code commits, right?

As I just said in #1991 (comment), in principle I am OK with it, as far as it does not flood our review capacity.

To me, the sad part is that there really is this much wrong with the code.

I will be focusing right now on going over the PRs. So far, everything I've seen makes sense and is good.

[rocky]

Thanks!