nx_dbg_step_gate.nx

nx_dbg_step_gate.nx -- the VERDICT gate for LN12 (nx_dbg_step, watch symbol dbg_step_over). THE ACCEPT RULE (/compare/lang LN12): break and step on our own binaries over the line table we emit. So this gate does not mock anything: * it ASSEMBLES a real RV64 program containing a real linking CALL, * it EMITS that program's .debug_line with nx_dwarf_line.nx -- OUR OWN DWARF v5 emitter, the one LB5 shipped and GNU readelf validates -- so the table under test is the table we produce, * it runs the deployed nx_dbg_step.elf against the pair and compares against expectations that the debugger had no part in computing. THE SUBJECT IS NEVER ITS OWN ORACLE. Every expected value comes from somewhere else: breakpoint address = the gate's OWN fixture layout (it placed line 12 at offset 8; the debugger has to rediscover that through the line table, and the two must agree). callee / caller line = likewise the gate's layout. exit code and halt = a SECOND, DEBUGGER-FREE sim (bootcap_machine) run in this process. never-executed addr = not assumed -- the reference sim records every pc it fetches and the gate PROVES the neg-control address is absent from that set before using it. the call is a call = decoded arithmetically with rv64im_min_decoder (kind == JAL, rd == ra) BEFORE any stepping outcome is asserted, so a fixture that could not exercise step-over cannot score a pass. WHY THE DISCRIMINATING TOOTH IS THE ONE THAT MATTERS. A debugger that single-steps INSTRUCTIONS passes "the line changed" trivially. The only thing that separates a stepping debugger from an instruction stepper is that STEP-OVER MUST NOT REPORT THE CALLEE'S LINE while STEP-INTO MUST. Both are asserted here, on the same fixture, from the same starting pc -- two tests that each isolate one signal would not prove discrimination between them. NEVER-BRICK IS TESTED BEHAVIOURALLY, NOT BY GREPPING FOR A WORD. A pid-shaped argument must be REFUSED (a process id can never become a subject), and the source must contain no sys_ptrace call site. The behavioural half cannot be fooled by prose, which a source scan can be -- this file and the organ both discuss ptrace at length, and a detector that finds its own explanation of the bug it hunts is the trap this estate has already paid for twice. Fixtures live in /tmp/nx_dbg_step_gate/ (created at setup, so a crashed run cannot leave a fixture beside a production beat) and are ASSEMBLED AT RUNTIME, never checked in. license_tier: ORIGINAL