1.5.6 Catch the broken formatters

Choose a few inputs that make four broken copies of a mismatch formatter give themselves away.

Hardware context: Regression diagnostics for a tagged bus or DMA transaction stream. Choose diagnostic inputs that separate a correct mismatch formatter from copies with planted formatting bugs.

A mismatch message is only useful if it is right. Your team's format_mismatch builds lines like ID=7 WIDTH=8 EXP=00x1z001 GOT=00x1z101 from a transaction ID, a width and two strings of bit symbols.

The version in the specification is correct. Four copies of it each hide one bug. Your job is to write the tests that expose them: in the task stimulus, call try_format once for each input you want to try. You may try at most three inputs.

For every input, the correct formatter and each broken copy run side by side. A broken copy is caught when its message differs from the correct one on at least one of your inputs. After each run you see which copies you caught and what each bug was.

Choose inputs that exercise every rule: the ID limits, the letter case of x and z, the string lengths, and values that differ.

Signals in the timeline

// Symbols are 0, 1, x, X, z and Z. X and Z are printed in lowercase.
function automatic bit symbolic_bits(input string text, input int width);
  if (text.len() != width) return 0;
  for (int i = 0; i < text.len(); i++)
    if (text[i] != "0" && text[i] != "1" && text[i] != "x" && text[i] != "X" && text[i] != "z" && text[i] != "Z") return 0;
  return 1;
endfunction

function automatic string lowercase(input string text);
  string out;
  out = "";
  for (int i = 0; i < text.len(); i++)
    out = $sformatf("%s%c", out, (text[i] == "X" || text[i] == "Z") ? text[i] + 8'd32 : text[i]);
  return out;
endfunction

function automatic string format_mismatch(input int id, width, input string expected_bits, observed_bits);
  if (id < 0 || id > 255) return "ERR:ID";
  if (width < 1 || width > 32) return "ERR:WIDTH";
  if (!symbolic_bits(expected_bits, width)) return "ERR:EXPECTED";
  if (!symbolic_bits(observed_bits, width)) return "ERR:OBSERVED";
  return $sformatf("ID=%0d WIDTH=%0d EXP=%s GOT=%s", id, width, lowercase(expected_bits), lowercase(observed_bits));
endfunction

task automatic stimulus();
  try_format(7, 8, "00x1z001", "00X1Z101");   // one call per input
endtask

Examples

Constraints and contract

• Call try_format at most 3 times inside stimulus.
• Each string argument is at most 40 characters.
• Do not call format_mismatch yourself; the comparison runs on every input you add.

Hints

Worked solution

Approach and reasoning

Each input in the reference stimulus is unusual in several ways at once. The first uses the largest legal ID, uppercase symbols and two different values, so it separates the copy that rejects ID 255, the copy that skips lowercasing and the copy that prints the expected value twice.

The second input is too short for its width. The correct formatter rejects it, and a copy that checks the string against its own length instead of width prints it.

Follow the example

1. Read the rules

   ID range, width range, string length and alphabet, lowercase output, error order.

2. Find an input per rule

   ID 255; uppercase X and Z; observed shorter than width; expected differs from observed.

3. Combine

   Two inputs cover all four rules within the limit of three.

The reference implementation is not published on this site.

Complexity

Each input costs one call of every design.

What the regression checks

5 directed, exhaustive or reproducibly generated checks cover the stated contract. The reference solution and deliberately broken variants are run through the same real SystemVerilog compiler during project validation.