Skip to content
Assembly and Fortran

Session record: 2026-10-06

This file is a chronological record of one working session between the owner and Claude Code. The session merged the review of the SNES video guides and built test 4, the first test that comes from that review. The decisions are in ../DECISIONS.md. The plan is in ../PLAN.md. The session before is in 2026-10-05.md.

Round 1: merge the review, and publish the site

Section titled “Round 1: merge the review, and publish the site”
  • What the owner asked: merge the pull request of the SNES review and make the website again.
  • Done: the pull request went into main. GitHub Actions built the game, ran all checks, and published the site. The review dashboard shows documents 13 to 15, and the segment pages show the new “Watch out” notes.

Round 2: test 4, the decimal counter, with its wrong twin

Section titled “Round 2: test 4, the decimal counter, with its wrong twin”
  • What the owner asked: build test 4, the decimal counter with its wrong twin. Make a passing version and a version that fails, as in tests 2 and 3. Show which part of the review and the documents leads to this test.
  • The A side: src/bcd.inc holds two routines that the game and the test share. bcd_inc adds 1 with add al, 1 and daa on the low byte, then adc al, 0 and daa on the high byte. After 9999 comes 0000, with CF = 1. bcd_digits changes each nibble into a character with XLAT. src/counter.asm shows four large digits from the 8 x 8 font of the BIOS, the 16 bits under them, and a carry cell. The key J jumps to 9990. COUNTER.COM is 623 bytes.
  • The shared keys: src/common.inc reads the key J only when the program sets JUMP_KEY. DOT3D.COM and SPIN3D.COM keep their sizes, and their checks still pass.
  • The test: test/checkbcd.asm runs bcd_inc 10,000 times and writes BCD.BIN: 10,000 values, the value after 9999, and the carry (20,003 bytes).
  • The B side: lab/check_bcd.f90 makes each value from the number of steps with MOD and ISHFT, and compares all of them. It also requires 0000 and a carry of 1 after 9999.
  • The result: the two sides agree on 10,000 of 10,000 values, and the step after 9999 is right.
  • The wrong twin: the switch -dWRONG_DECIMAL removes the two DAA words. The counter then counts in binary: 0009, 000A, 000B. The check finds 9,990 wrong values, the number that PLAN.md gave before the code existed. After 9999 steps the counter is 2710h with no carry. build.sh requires this check to fail.
  • The pages: /game/test4a/ and /game/test4b/, with the A and B panels. The panels read the 16 bit cells and the carry cell from the screen of the emulator, so they follow the counter. In Chrome, the panels followed the counter, the key J went to 9990, and the counter went from 9999 to 0000. The page /game/ lists the two new tests.
  • The path from the review: the page of test 4a has the section “Where this test comes from”. It links the moment in the video (32:05), the findings D13-M11, D15-M7, D14-M1 and D15-M1, learning unit U3 in the summary, the segment pages that teach DAA, the design in PLAN.md (Phase 4d), and the dictionary words. The page of test 4b links the rule of the wrong twin.
  • A note for later: on the game pages, the first click on a button sometimes does nothing, and the browser reports that simulateKeyPress is not a function. The pages of test 3 do the same. The next clicks work. This change does not fix it.
  • Decision: D27.