interpreter: Report correct number of targets in subprojects

[dnicolodi]

Fixes #16116

[dnicolodi]

I'm not so sure this makes the reported number of targets easier to grok. With this, each subprojects reports the number of targers in itself and all of its subprojects, however, the log output is not super clear regarding the tree structure of the subprojects. Maybe this needs some more thinking.

[dcbaker]

Could we write this as something like len(x for x in self.build.targets if x.subproject == self.subproject and x.for_machine == self.for_machine) (Can't remember exactly which attributes we have here)

[dnicolodi]

Sure, but the interpreter does not have a for_machine attribute for the machine type of the subproject being processed. The only way I see of getting to it is to use the last entry in self.subproject_stack and special handling the root project. Or should I add Interpreter.for_machine?

[bonzini]

Given in the future there could be more ways to invoke the subproject (depending on how Meson introduces the possibility to invoke subprojects with different machine maps) that approach seems more complex than the subtraction here. Another possibility is to add a variable with the count and adjust it in add_target; but this PR looks good to me, if it's decided this was a bug and not just a badly worded message.

[dnicolodi]

I am not happy with the count reported with the subtraction approach. This approach still reports the current subproject targets plus all the targets in subprojects called from it. This is a bit less confusing that the cumulative number, but still not great. The approach suggested by @dcbaker would give the least surprising number.

Another possibility is to add a variable with the count and adjust it in add_target

I am not sure this would help. The problem would then become when to reset this new variable to 0. Subprojects are organised as a tree, thus the count would need to be stored as a stack, similarly to Intepreter.subproject_stack (or an entry would need to be added to this stack).

Given in the future there could be more ways to invoke the subproject (depending on how Meson introduces the possibility to invoke subprojects with different machine maps) that approach seems more complex than the subtraction here.

Wouldn't this require to switch from storing the for_machine information to storing the machine_map information in the targets? If so, the count problem would be solved filtering for (subproject, machine_map) instead than for (subproject, for_machine). But I am not familiar with this code at all, thus I may be missing something.

[bonzini]

The problem would then become when to reset this new variable to 0.

Each Interpreter would have its own copy, starting at zero.

[dnicolodi]

I eeworked it in this way. PTAL