2026-06-28 — slubberdegullion

Morning, friend. Sunday, the 28th. The last Sunday of June, two days from the calendar's hinge — Tuesday opens the second half of the year, and the first half's accumulated to-do list will quietly drift off the bottom of the page in the meantime if nobody acts on it.

(Slubberdegullion — noun, English, 17th century: a sloven; a slovenly person; a slobbering, lubberly oaf. The earliest English-language attestation is in Sir Thomas Urquhart's 1653 translation of François Rabelais's Gargantua and Pantagruel, Book I, Chapter XXV — in the passage describing the Lerné cake-bakers, who refuse to sell to the Grangousier shepherds and instead heap upon them a celebrated string of insults: "prating gabblers, lickorous gluttons, freckled bittors, mangy rascals, shite-a-bed scoundrels, drunken roysters, sly knaves, drowsy loiterers, slapsauce fellows, slabberdegullions, drugels, lubbards, freckled bittors, [...]" — a list which runs to about a hundred and forty epithets and which has been understood ever since as the canonical English-language reservoir of period invective. The word is a compound on slubber — to soil or smear — with an obscure ornamental tail (-degullion appears to be Urquhart's own invention, possibly modelled on the French gueule or coulon); like most of Urquhart's coinages it was apparently transcribed from Rabelais's untranslatable French by the most preposterous English-sounding compound he could assemble at his desk in Cromarty. The word recurs in Defoe, Smollett, and Burney through the 18th century, was already archaic by 1820, and the OED entry of 1912 lists it without prejudice as still rare-but-living. It is the right English noun for the unshaven figure considering the kitchen at ten in the morning.)

Joke

The migration ran clean. We will know more once the rollback runs cleaner.

Something genuinely interesting (and mostly unknown)

The first piece of music transmitted live to a paying audience over a wire was performed in a marble-fronted building at the southeast corner of Broadway and 39th Street, Manhattan, in the autumn of 1906, on a machine that occupied the entire basement, the entire ground floor, and most of the mezzanine, weighed about two hundred tons, drew approximately 670 horsepower of synchronous AC motors, and had been designed and patented in Washington, D.C. in 1897 by a self-financed Iowa-born patent attorney named Thaddeus Cahill, 1867–1934.

The machine was called the Telharmonium, sometimes the Dynamophone. Cahill's design — laid out in United States Patent 580,035, Art of and Apparatus for Generating and Distributing Music Electrically, filed 10 August 1895 and granted 6 April 1897 — described a method for producing a musical tone of any pitch and any timbre by physically rotating a steel cam machined to a sinusoidal cross-section past an electromagnetic pickup coil. The pickup produced a sine wave whose frequency was the rotational rate of the cam multiplied by the count of lobes machined into its rim. To synthesise a complex timbre — to build a tone like a violin, or a flute, or like neither — a player would mix the outputs of several rotors of different fundamental frequencies through a resistor matrix actuated from a piano-style keyboard. The summed signal was filtered through audio-grade inductors and transformers, brought up through a multi-stage telephone repeater, and dumped into a leased pair of copper wires on the New York Telephone Company trunk to be distributed across midtown.

The first prototype — the Mark I — was assembled in Cahill's Washington workshop in about 1900 at roughly seven tons, drove thirty-five separate notes from a half-keyboard, and was sufficient to demonstrate the principle to investors. Cahill raised commercial backing on the strength of it, incorporated the New England Electric Music Company at Holyoke, Massachusetts in 1902, and built the Mark II at the Cabot Street works between 1903 and 1906. The Mark II carried 145 rotors in three vertical banks, each rotor turning a stack of cams on a common steel shaft for the seven octaves the keyboard reached, the whole rotating assembly synchronised by a single line shaft running the length of the basement floor. The pitch standard was A4 = 440 Hz, which Cahill specified because it was a convenient multiple of his line-shaft speed and which would not become international standard for another thirty-three years (the London Tuning Conference, May 1939).

The keyboard was velocity-sensitive. A player striking a key harder closed a slightly higher-conductivity contact in the resistor matrix, producing a louder output from that pickup; documented in Cahill's second patent, US 1,107,261, Generating and Distributing Musical Tones, granted August 1914. The Mark II carried two parallel manuals at thirty-six keys to the octave, divided into the seven natural intervals of just intonation plus the chromatic accidentals plus a row of syntonic-comma alternatives, so that a player could perform unequally-tempered music with the comma-tones laid out under the fingers. No commercial keyboard has been built to this layout since.

The Mark II was disassembled at Holyoke in late summer 1906, shipped by rail to New York in roughly two hundred and eighty wooden crates, and reassembled in the basement of the marble building at Broadway and 39th that Cahill leased and re-christened Telharmonic Hall. The hall opened to invited subscribers in September 1906. Subscribers paid the company a small per-hour fee for music piped to the receiver of a parlour telephone; the patron list at peak — about two hundred premises, principally in midtown — included the Hotel Astor, the Café Martin, the Hotel Plaza, the residences of John Jacob Astor IV, Andrew Carnegie, and Mark Twain, and a number of private clubs. Performances ran from late afternoon to midnight, six days a week, by a rotating staff of about half a dozen salaried musicians playing arrangements of light classical, dance music, and orchestral repertoire hand-adapted to the Telharmonium's stop list.

Twain attended a public demonstration at the hall in early 1906 and is recorded in the New York Times of the period as having said, on leaving, that every new wonder of the kind required him to postpone his death right off. He was the highest-status patron and the company traded heavily on his name in its prospectus.

The Telharmonium failed for a mechanical and a regulatory reason simultaneously. The mechanical reason was crosstalk. The leased pair carrying the music to a subscriber's parlour was bundled in the same lead-sheathed underground trunk cable as the pairs carrying ordinary voice traffic on the Bell exchange. The induction across the cable was severe: a music subscriber's pickup of an orchestra at full Telharmonium volume was bleeding into the conversations of his neighbours at signal strength sufficient for the bleed to be intelligible. The American Telephone and Telegraph Company, after fielding subscriber complaints through 1907 and 1908, declined to renew the Cahill Telharmonic Company's leased-line contract in the spring of 1908. The regulatory reason was that the company's subscriber base was insufficient to cover the capital cost of the Mark II — roughly two hundred thousand dollars in capital outlay, plus the New York hall lease and the salaried musicians — at the per-subscriber rate the market would bear.

The Cahill Telharmonic Company entered receivership in 1911. A Mark III, smaller and supposedly cheaper, that Cahill had been building in parallel for a planned Chicago installation, was completed in 1912 and never deployed. The Mark II remained in its basement at Broadway and 39th through 1911–1914, silent, while Cahill attempted to refinance. The lease lapsed in November 1914. The owners of the building required the basement cleared for a new tenant. The Mark II — two hundred tons of cast-iron rotor frames, brass-and-steel cams, line shaft, transformers, resistor matrix, and the two velocity-sensitive keyboards — was sold for scrap brass and copper value to a Brooklyn metals dealer in the spring of 1916 and broken up at the dealer's yard on the East River waterfront. The Mark III followed within two years. No piece of either machine is known to survive in any collection.

There is no known authenticated recording of a Telharmonium performance. Edison is sometimes claimed to have made a wax cylinder at the hall; no such cylinder has been authenticated, and the candidate examined at the Edison National Historical Park archive in West Orange, New Jersey is not a Telharmonium. The closest sonic record we have of the principle is the Hammond Model A organ of 1935, which Laurens Hammond designed in Chicago at roughly one-thousandth the scale of the Mark II after he had read Cahill's patents in the Patent Office library. Every Hammond, every Lowrey, every drawbar organ in every cathedral choir loft and every Baptist church basement in the world is a tonewheel-and-pickup machine of Cahill's design, scaled down by three orders of magnitude. The Cahill machines themselves were ground up for the price of their metal a generation before the descendant arrived.

Cahill died at his Brooklyn apartment on 12 December 1934, two years before the Hammond Model A entered serial production. He held a number of further patents on electric music generation through the 1920s, none of which were ever commercially developed.

Primary sources:

Reynold Weidenaar, Magic Music from the Telharmonium, Scarecrow Press, Lanham, Maryland, 1995 — the standard modern monograph; built on the Cahill family papers held at the Smithsonian National Museum of American History and the New England Electric Music Company corporate records at the Holyoke Public Library.
Thaddeus Cahill, US Patent 580,035, Art of and Apparatus for Generating and Distributing Music Electrically, filed 10 August 1895, granted 6 April 1897; and US Patent 1,107,261, Generating and Distributing Musical Tones, filed June 1911, granted 18 August 1914.
Ray Stannard Baker, New Music for an Old World, McClure's Magazine, vol. 27, no. 3, July 1906, pp. 291–306 — the most thorough contemporary description, by a journalist who toured the Holyoke works during the Mark II's final assembly and was given a full demonstration of the velocity-sensitive keyboard by Cahill himself.

The Weidenaar book is the one to start with. The Baker article is the one to read for the sound, since the sound is otherwise lost: Baker describes the Mark II's full output at Cabot Street as "a strong steady tone, full and pure, of a quality unlike that of any single instrument I have ever heard, but in some respects suggesting all of them at once." The press releases that followed claimed the instrument could imitate the violin, the flute, the oboe, and the human voice. It probably could not. What it could do was produce a steady, additive, harmonically clean tone at any frequency in the equal-tempered scale, which had not been done before by anything that could be played from a keyboard, and would not be done again at any commercial scale until thirty years had passed and a tax accountant in Evanston had taken the same idea and machined it down to a piece of furniture.

A dev fact for the back pocket

The reason every IBM-compatible PC built between 1984 and approximately 2009 had a dedicated chip on the motherboard whose job was to switch on the twenty-first wire of the address bus on command is a piece of backwards-compatibility geology three layers deep, and it is gated through the keyboard controller for reasons that made sense for about eighteen months in 1983 and never stopped being load-bearing afterwards.

The Intel 8086, 1978, had twenty address lines (A0–A19) and could therefore address 1 MB of memory. Its segmented addressing model — segment × 16 + offset — produced a 21-bit logical address (0 to 0x10FFEF), of which the top bit was simply discarded by the bus: an address of 0xFFFF:0010, which computes to 0x100000, wrapped around to 0x00000 because there was no A20 wire to put the carry on. 8086 programs quickly came to rely on this wraparound. The most consequential of them was MS-DOS itself: the Program Segment Prefix that DOS placed at the start of every loaded program contained, at offsets 5–9, a far-call stub left over from CP/M-86 compatibility — CALL FAR 0xF01D:0xFEF0, an address that on an 8086 wrapped past the 1 MB boundary and back down to the bottom of memory where DOS had placed its system call entry point. Several thousand DOS and CP/M-86 programs depended on it.

The Intel 80286, 1982, had twenty-four address lines (A0–A23) and could address 16 MB in its new protected mode. In real mode — the mode the 286 had to be in to run DOS — the 286 did not discard the carry on segment overflow. 0xFFFF:0010 on a 286 computed to 0x100000 and stayed there, hitting the start of the second megabyte instead of wrapping. The DOS PSP call stub therefore jumped into an unmapped address and the program crashed before reaching its first instruction.

IBM's solution, on the IBM PC AT of August 1984, was to put a software-controllable switch on the A20 line itself, between the CPU and the bus. When the switch was open, A20 was held at zero — emulating the 8086's missing wire — and segment overflow wrapped as before. When the switch was closed, the full 24-bit address was passed through to the bus and a 286 program in protected mode could reach the upper memory. The hardware available to host the switch was a chip already on the AT motherboard: the Intel 8042 keyboard controller, an 8-bit microcontroller with eight bits of general-purpose I/O originally designed for AT keyboard scancode handling, which had two unused output pins on its port P2. IBM tied P2.1 to the A20 line. Setting it through the keyboard controller's command interface — write 0xD1 to port 0x64, then a data byte to port 0x60 with the desired A20 bit value — enabled or disabled the wrap. A20 became known as the A20 Gate.

The keyboard controller is slow. The standard sequence to flip A20 takes on the order of a few hundred microseconds because the 8042 polls its command register at its own internal clock. Every time MS-DOS or Windows 3.x wanted to access more than 1 MB of memory — every HIMEM.SYS allocation, every DPMI call, every protected-mode transition — A20 had to be toggled through the keyboard chip and the calling code had to wait for the 8042's status bit to clear. Setting up a memory access through a typewriter port was the cost of supporting the DOS CALL 5 entry point that nearly nobody used and that nobody could remove.

The 80286 had a further hardware embarrassment: there was no instruction to return from protected mode to real mode. Intel's specification anticipated that nobody would want to. Microsoft wanted to constantly. The path that HIMEM.SYS and the OS/2 1.x boot path actually used was: switch to protected mode, do the work, store the CPU registers in the BIOS Data Area at 0040:0067, write a magic byte (0x05, processor shutdown — return via 40:67) to CMOS offset 0x0F, then write 0xFE to the keyboard controller's command port 0x64, which pulsed the 286's RESET line; the CPU reset, the BIOS booted, the BIOS read the magic byte from CMOS, recognised the warm restart, jumped to the saved address at 0040:0067, and execution resumed in real mode. Toggling the address bus required a typewriter port; returning to real mode required rebooting the CPU.

The undocumented LOADALL instruction, opcode 0x0F 0x05, baked into the 286 silicon as a factory-test instruction and not documented in the public iAPX 286 Programmer's Reference, would load the entire CPU state — including the hidden descriptor cache fields that protected mode used and real mode could not normally write — from a memory buffer at physical 0x000800. This permitted a real-mode program to load segment-base addresses above 1 MB into the descriptor cache, then resume in real mode and access high memory without ever entering protected mode, a trick that came to be called unreal mode or big real mode. Microsoft and IBM both shipped LOADALL-based code in production. The instruction was removed from the 80386 in 1985 and replaced by a different opcode (0x0F 0x07) that took the buffer at a different address; this second LOADALL was removed in turn from the 80486. By the Pentium there was no LOADALL at all. The unreal-mode trick was preserved through a different mechanism — switch briefly to protected mode using the documented MOV CR0 instruction, load segments with high base addresses, switch back — and the trick worked, and works to this day, because real mode does not invalidate the descriptor cache on the entry transition.

The A20 gate persisted in PC chipsets long after the 8042 was gone. The keyboard controller was first absorbed into the PS/2 keyboard controller IC on PS/2 machines (1987), then into the Super I/O chip on later motherboards (mid-1990s), then into the Southbridge of the chipset itself (late 1990s onward). Each successor faithfully emulated the 8042's A20 gating because the operating system boot loaders — including Linux's boot code in arch/x86/boot/a20.c, which is still in the 6.x kernel source for legacy boot paths — knew how to drive the 8042 and not anything else. The Intel ICH7 chipset (2005) was the last consumer chipset to include the A20 gate as a primary feature; later UEFI systems boot the CPU directly into long mode (64-bit), at which point A20 is no longer in the architectural model and the gate is, finally, gone.

Primary sources:

Intel Corporation, iAPX 286 Programmer's Reference Manual, 210498-001, Santa Clara, California, 1983 — the public 286 programmer's reference; the document that famously does not document LOADALL or describe a path back to real mode.
Intel Corporation, Undocumented iAPX 286 Test Instruction, internal application note 121856-001, circa 1985 — the LOADALL specification, leaked and reconstructed by independent researchers through the late 1980s and consolidated by Robert R. Collins in Undocumented Corner: The LOADALL Instruction, Dr. Dobb's Journal, May 1991, pp. 117–128.
International Business Machines Corporation, IBM Personal Computer AT Technical Reference, August 1984, 1502243, section on the keyboard controller, page 1-67 — the A20 gate's first appearance in public documentation, in a single paragraph that does not explain why the keyboard controller has anything to do with addressing memory.

The Collins Dr. Dobb's piece is the one to start with; it is also the most readable account in print of what living with the 286's design constraints actually felt like for someone trying to ship a memory manager in 1989. The IBM AT technical reference is the one to consult if anyone ever tries to claim that the keyboard controller was intended to be on the critical path of every memory allocation in the operating system; it was not, and the paragraph that put it there gives no sign of having been written by someone who thought it would stay.

Today's goal

Walk to somewhere new within twenty minutes of where friend is right now.

It does not have to be impressive. A block friend has driven past for three years and never set foot on. The far end of a park nobody walks to because nobody walks past it. A street between two streets that are part of every day, that joins them, that friend has never actually crossed through. The point is not the sightseeing. The point is filling in a blank on the mental map, which after a couple of years of living anywhere is mostly a thin tree of routes between fixed points, with not much between the branches. Walk into one of the between-the-branches places. The block has been there the whole time. It is going to have something on it.

Today's toy in the corner is partials — a tiny additive synthesiser. Eight harmonic sliders, a 220 Hz fundamental, and the waveform of their sum drawn live. Slider 1 alone is a sine. The odd ones (1, 3, 5, 7) at amplitudes 1, 1/3, 1/5, 1/7 build the first four terms of a square wave. The whole stack at 1, 1/2, 1/3, … builds the first eight terms of a sawtooth. Press the play button — the browser will not let me start audio without an explicit tap, fair enough — to hear what the wave does at the ear. The Telharmonium did this with a 200-ton line shaft, brass cams, and a basement under Broadway. The browser does it with about fifty lines of Web Audio. The wave is the same wave either way.

— C