write_blif: name CIs after the node, not the topological index

[marcelwa]

write_blif names combinational inputs pi{topo_ntk.node_to_index(n)}, while every other reference in the same writer — fanin lists and the PO bridges — names a node by its raw id. topo_view reimplements node_to_index as the position in the topological order, so the two agree only for as long as the CI node ids happen to be contiguous.

They are not contiguous in general. A klut_network produced by lut_map has gaps in them, and so does any network where a primary input is created after a gate.

What it produces

The smallest case — two PIs, an AND, then a third PI:

klut_network klut;
auto const a = klut.create_pi();
auto const b = klut.create_pi();
auto const f = klut.create_and( a, b );
auto const c = klut.create_pi();       // CI ids are now 2, 3, 5
klut.create_po( klut.create_and( f, c ) );

.model top
.inputs pi2 pi3 pi4        <- pi4 is declared and never read
.outputs po0
...
.names new_n4 pi5 new_n6   <- pi5 is read and never declared

Why it is worth fixing rather than working around

The output is not malformed in a way anything reports. lorina::read_blif does reject it, but ABC accepts it and ties the undeclared signal to constant 0 — so the file reads back as a well-formed circuit that computes a different function, silently, with the right port count.

I hit this running combinational equivalence checking over LUT-mapped EPFL circuits: mem_ctrl came back NOT_EQUIVALENT with 39 phantom inputs, and it took a while to believe the writer rather than the mapper.

The change

Two lines: derive the name from the node, which is what the rest of the writer already does. The named-network branch has the same problem — it calls make_signal(node_to_index(n)), so has_name/get_name are looked up against the wrong signal — and is fixed the same way.

The added test builds the smallest network with a CI gap and pins the exact output. It fails on master (both the string comparison and the blif_read_after_write_test on top of it) and passes with the change; the rest of the suite is unaffected — 917 test cases, 133227 assertions green.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.06%. Comparing base (852605f) to head (84fc452).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #704      +/-   ##
==========================================
- Coverage   84.07%   84.06%   -0.02%     
==========================================
  Files         190      190              
  Lines       29513    29513              
==========================================
- Hits        24813    24810       -3     
- Misses       4700     4703       +3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.