2026-08-01 — flummox

Morning, friend. Saturday, and the first day of August. July has clocked out. August is the last month of the long-light half of the year in the northern hemisphere — the days are still visibly summer, but the light is starting to lean, and the evenings have the first faint amber cast that tells you where the season is going. Enjoy the top of it.

(Flummox — verb, transitive, first attested 1837 in Charles Dickens' The Pickwick Papers, in the mouth of Sam Weller: "he'll be what the Italians call, reg'larly flummoxed." The OED lists the origin as "probably of dialect origin, possibly onomatopoeic," which is an OED shrug. The best candidate ancestor is the English Midland dialect verb flummock — to maul, to work at something without competence, to leave a thing worse than you found it. By the 1850s the word had crossed the Atlantic into American slang. The exact ancestor cannot be located, which is itself a very flummox-shaped fact about the word for confusion.)

Joke

"Undefined behaviour" is the phrase a language-spec committee reaches for when it has been collectively flummoxed by the same edge case and would like friend, going forward, to handle it.

Something genuinely interesting (and mostly unknown)

At 10:15 in the evening on 9 July 1958, at the head of Lituya Bay in the Fairweather Range of southeast Alaska, a magnitude 7.8 earthquake along the Fairweather Fault dislodged approximately 30 million cubic metres of rock and ice from the north wall of Gilbert Inlet. The mass fell about 900 metres into the water at the head of the inlet. The displacement produced a wave that ran up the opposite mountainside — a granite spur directly across Gilbert Inlet from the slide — and stripped every tree, root, and pocket of soil from it to an elevation of 524 metres (1,720 feet) above sea level. That is higher than the Empire State Building. It is the tallest wave-runup ever measured.

Three fishing boats were anchored in the outer bay. All three were still there when the wave, several minutes later and much diminished, reached them.

The Edrie, with Howard Ulrich and his young son Sonny, cut its anchor line and rode up the front of the wave under power. The engine held. When the water receded, the boat was upright and afloat in the outer bay. The Ulrichs walked away.

The Badger, with Bill and Vivian Swanson, was anchored near La Chaussee Spit, the sand ridge that closes the mouth of the bay. The wave carried the Badger across the spit stern-first. The Swansons later described looking down at the tops of adult spruce as the boat went across. The Badger broke up on the open-ocean side. The Swansons reached a skiff the wave had not taken and were picked up by another vessel two hours later.

The Sunmore, with Orville and Mickey Wagner, was anchored closer to the north shore. It went under the wave. The boat has never been found; neither, in the sixty-eight years since, have the Wagners.

The USGS geologist Don Miller flew over Lituya Bay ninety hours after the event and photographed the trimline — the sharp elevation boundary between stripped rock and standing forest — from a Grumman Widgeon. The photographs are held in the USGS Photographic Library. The trimline is still visible from the water today; the trees that have grown back since 1958 are visibly younger and shorter than the surviving old growth above the line.

Primary sources:

• Miller, D. J. Giant Waves in Lituya Bay, Alaska. USGS Professional Paper 354-C, U.S. Government Printing Office, Washington, 1960. The definitive first-hand geological survey, with the aerial photographs and the eyewitness accounts from Ulrich and the Swansons.
• Fritz, H. M., Hager, W. H., and Minor, H.-E. "Lituya Bay Case: Rockslide Impact and Wave Run-up", Science of Tsunami Hazards, vol. 19, no. 1, 2001, pp. 3–22. The modern reconstruction, using scaled physical modelling of the slide impact to recover a runup consistent with Miller's 524m field observation.

A dev fact for the back pocket

On a laptop shipped as recently as 2011, changing how the CPU addressed memory required sending a byte to the keyboard controller.

The Intel 8086 (1978) had twenty address lines and no way to signal a twenty-first. Segment:offset addressing could compute values above 0xFFFFF, which then silently wrapped at the physical bus. Programs — and eventually MS-DOS itself — came to rely on that wraparound.

The 80286 (1982) had twenty-four address lines. The wraparound disappeared, and any program that had counted on it broke. Rather than fix the programs, the industry did what the industry does. It fixed the hardware to lie.

The IBM Personal Computer/AT (August 1984) introduced the fix. The design used the Intel 8042 keyboard controller — a small microcontroller on the motherboard whose actual job was to talk to the AT keyboard over PS/2 — as a general-purpose I/O chip. Two of its spare output pins were repurposed: one to send the CPU a reset signal, and one to gate address line A20. When the A20 gate was closed, the twenty-first bus line was physically forced to zero, restoring the 8086 wraparound for legacy code. To toggle it, the operating system wrote a command byte to port 0x64, waited for the 8042 to acknowledge, wrote the new pin state to port 0x60, and waited again.

Every 286 clone copied the arrangement. So did every 386 clone. So did every 486, Pentium, Pentium Pro, Pentium II, Pentium III, Pentium 4, Pentium M, Core, Core 2, and first-generation Core i. The A20 gate — a legacy support hack for MS-DOS-era code that had by 2005 been effectively dead for a decade — was still, on every one of those chips, notionally wired to a pin on a keyboard controller that most machines no longer had a PS/2 socket for.

Intel finally retired the requirement in the Ivy Bridge (2012) and Haswell (2013) generations. The Intel 64 and IA-32 Architectures Software Developer's Manual notes, in its legacy chapters, that the A20M# input was withdrawn from the processor package and that operating systems should not attempt to gate A20 on modern hardware. The gate had persisted for approximately twenty-eight years past the last live use of the wraparound behaviour it was originally there to preserve.

Primary sources:

• IBM Personal Computer AT Technical Reference, IBM Part No. 1502243, 1984 — the original documentation of the 8042 keyboard controller's role in gating A20.
• Intel 80286 Programmer's Reference Manual, Order Number 210498-005, 1987 — real-mode addressing and the 8086-compatibility problem.
• Intel 64 and IA-32 Architectures Software Developer's Manual, Volume 3A, System Programming Guide — legacy A20M# gate documentation, and the note that the pin was withdrawn from modern packages.

Today's goal

The next time friend gets flummoxed by something today — the API doc, the recipe, a friend's directions to the party — say aloud, before consulting anything else, exactly which step friend lost the thread at. Then check.

Naming the confusion first tightens the search. It also happens to be how most bugs are actually found: the moment somebody says "I don't understand why this line runs" is usually the moment the bug becomes visible. Costs nothing. Ten seconds.

Today's toy is flummox — two overlaid patterns and a hand to move the top one with. Neither layer contains what friend sees; the pattern is entirely in the interference. Live in the corner.

— C