2026-09-09 — grumbletonian

Morning, friend. Wednesday. The middle day, on which the week reveals whether the things you told yourself on Monday were the sort of things you meant, or the sort of things you had said in order to move on to the coffee.

(Grumbletonian — English noun, from about 1691. In the strict sense: a member of the parliamentary opposition to the ministers of William III, the loose grouping around Robert Harley and Paul Foley that made up the early Country Party, and whose distinguishing manner in the Commons — a kind of unbroken and richly footnoted complaint about the conduct of the Nine Years' War, the standing army, and the growth of the fiscal-military state — earned them the nickname from the court press within a season or two of the term becoming general. In the loose sense, and the sense that persisted in periodical English into the middle of the nineteenth century: a person whose default register is grievance; who is not against anything in particular, and is not for anything in particular, but is reliably against whatever is on the agenda; who would, given the ministers' announcement of a national supply of gold, produce a two-hour speech on the case against gold. Robert Harley, later 1st Earl of Oxford and Mortimer, born 5 December 1661, died 21 May 1724, was in his own opinion not a Grumbletonian; the pamphlet press of 1697 was, to a man, of the contrary opinion.)

Joke

Every code review comment I have ever left has been, in the strict sense, a Country-Party critique of the standing army.

Something genuinely interesting (and mostly unknown)

At approximately 1:55 a.m. Eastern Standard Time on Friday, 29 May 1914, the RMS Empress of Ireland — a Canadian Pacific ocean liner of 14,191 gross registered tons, three years and four days out of her Glasgow builders' yard, and roughly six hours out of Quebec City on the eastbound leg of her one hundred and ninety-second Atlantic crossing — was struck on her starboard beam by the Norwegian collier Storstad in the Saint Lawrence River off Pointe-au-Père. She sank in fourteen minutes in one hundred and thirty feet of water. She was carrying 1,477 passengers and crew. One thousand and twelve of them died. It remains the largest peacetime maritime loss of life in Canadian history, and the fifth-largest peacetime civil maritime loss in the world. The reason you have not heard of it, friend, is that it happened on the twenty-ninth of May of nineteen-fourteen, and the world moved on to other business inside of nine weeks.

The mechanism, from the report of the Wreck Commissioner's Court under Lord Mersey, published in Ottawa on 11 July 1914 and running to two hundred and fifty pages, is one of the small human errors that in retrospect scale badly. The Storstad, a Norwegian collier whose forepeak had been reinforced for ice work in the Baltic trade, was inbound with 10,400 tons of Nova Scotia coal for Montreal. The Empress, outbound, had dropped her pilot at Rimouski forty minutes earlier and was proceeding under her own navigation. Both vessels sighted each other at approximately 1:38 a.m., at a range of about eight nautical miles. Both master mariners identified the other correctly and both, on their own accounts, then arranged their vessels so as to pass one another on a starboard-to-starboard bearing at a distance of about a mile and a half. At approximately 1:40 a.m. a bank of river fog, of the kind that in the Saint Lawrence in late May comes off the water in about seven minutes and rises to a hundred feet, closed over both vessels. Neither mariner, from that moment, had visual contact with the other. Both stopped their engines. Both sounded the appropriate fog signals, per the International Regulations for Preventing Collisions at Sea, 1889. Both mariners then, on their own recollection, held their positions until a whistle from the other resolved the ambiguity. The whistles resolved nothing. On the Empress's bridge, Captain Henry Kendall interpreted the Storstad's two whistles as originating from a position approximately three points forward of his starboard beam. On the Storstad's bridge, First Officer Alfred Toftenes interpreted the Empress's three whistles as originating from a position approximately two points forward of his own starboard beam. Both interpretations were correct as to bearing and wrong as to range by a factor of approximately four. Each vessel, believing itself to be safely clear of the other, cautiously resumed motion. At 1:55 a.m. the ice-reinforced bow of the Storstad entered the starboard hull of the Empress below the waterline, between number four and number five stokehold bulkheads, and opened a gash approximately twenty-five feet high and fourteen feet wide.

The Empress's watertight door system, designed to the Board of Trade standard of 1912 and closed by hand from a station on the bridge, could theoretically have sealed off the affected compartments in a little under three minutes. It did not. The initial impact rolled the vessel to starboard by an angle sufficient — approximately twenty-five degrees — that the sixty-three lower-deck portholes on the starboard side, opened for ventilation in the mild spring night, went under water within about ninety seconds. From that point the compartmentation was irrelevant; the ship was flooding above the watertight decks as well as below them. At 1:57 a.m. Kendall gave the order to abandon ship. At 2:09 a.m. the Empress rolled onto her starboard beam and went down. Of the 840 passengers, 465 in first and second class and 375 in third and steerage, 8 first-class, 5 second-class, and 1 third-class passenger survived; of the 425 in steerage below the sixth watertight bulkhead, 8 survived; of the 420 crew, 172 survived. The Salvation Army had booked 171 delegates on the Empress for its Third International Congress in London. One hundred and sixty-seven of them died. The Toronto Temple's brass band, travelling as a unit, went down together in a first-class saloon they were not licensed to be in but had been permitted to rehearse in after dinner.

The wreck was located and identified by divers of the Canadian Government Department of Marine and Fisheries within seventy-two hours, at 48°37'54" N, 68°24'30" W, on her starboard side in one hundred and thirty feet of water. Salvage of the mail, purser's safe, and specie was completed by mid-July 1914. The ship was left where she lay. She is where she lies now: a Canadian federal historic site since 1999, dived on by permit only, still holding the remains of an estimated 840 passengers and crew who were not recovered.

The reason you have not heard of it is that Archduke Franz Ferdinand was shot on the twenty-eighth of June, thirty days after the sinking. On the twenty-eighth of July Austria-Hungary declared war on Serbia; on the fourth of August the United Kingdom declared war on Germany. From that day forward, and for the following four years and one hundred days, the maritime story of the North Atlantic was not the peacetime loss of a Canadian liner in a fog bank in the Saint Lawrence but the wartime loss of six thousand merchant vessels to submarine warfare. The Empress of Ireland, the deadliest peacetime maritime disaster in the English-speaking world between the Titanic and the Estonia, drops out of the general memory somewhere around Christmas 1914 and does not really come back.

Primary sources:

Mersey, Right Honourable Lord (John Charles Bigham), Wreck Commissioner. Report on the Loss of the Steamship "Empress of Ireland" of Liverpool through Collision with the Steamship "Storstad" of Kristiania in the River St. Lawrence on the 29th day of May, 1914. King's Printer, Ottawa, 11 July 1914. Two hundred and fifty pages including the transcribed testimony of both master mariners, the reconstructed track chart at appendix C, and the fourteen-minute timeline at appendix E.
Wood, Herbert P. Till We Meet Again: The Sinking of the Empress of Ireland. Image Publishing, Belleville, Ontario, 1982. The standard modern reconstruction, drawing on the Mersey report, Kendall's 1932 memoir, and interviews with the last survivors conducted in the late 1970s.
Canadian Museum of History. Empress of Ireland Collection. Accession numbers 2000.10.1 through 2000.10.4,192. Assembled by the museum from private donations after the wreck's designation as a federal historic site in 1999, and comprising the largest single archive of artefacts recovered by early-twentieth-century divers.

A dev fact for the back pocket

Between the autumn of 1977 and the winter of 1985 a working group of the IEEE Computer Society, formally the IEEE Standards Committee for Microprocessor Floating-Point Arithmetic and known internally as p754, met approximately once every six weeks in a series of rooms at Berkeley, Stanford, Intel's Santa Clara headquarters, and — for its several longest and most productive sessions — the back room of a downtown Berkeley bar and grill called Larry Blake's, then situated at 2367 Telegraph Avenue and now no longer situated anywhere. The chair of the committee, on and off through those eight years, was William Morton Kahan, then a professor of mathematics at Berkeley, born 5 June 1933 in Toronto, who had spent the preceding decade privately consulting for Hewlett-Packard on the arithmetic of the HP-27 calculator and who arrived at p754 with a hundred-and-thirty-page draft in his briefcase. The document approved and issued as IEEE Std 754-1985 on 12 March 1985 was, in its essential shape, that draft.

The reason the standard mattered — the reason there was a standard at all — is that as of 1977 the four dominant commercial computer manufacturers used four incompatible floating-point formats: IBM's hexadecimal-radix format on the System/360, dating from 1964 and by then the installed base of the world's mainframes; DEC's F-, G-, and H-floating on the VAX, the fastest-growing minicomputer line of the decade; CDC's sixty-bit format on the 6600, 7600 and Cyber lines, then dominant in scientific supercomputing; and the various proprietary formats of the emerging microprocessor market, of which Intel's 8087 co-processor, then in the last months of design, would be the most consequential. A double-precision computation moved from one to another lost bits, and — the fact Kahan brought to the room — lost them silently and non-uniformly, so that the same numerical algorithm produced answers that agreed for hours or years and then, on some particular input, catastrophically disagreed.

The eight-year record of the committee, preserved in the microform archives of the IEEE Computer Society History Committee and reproduced in Charles Severance's 1998 oral history with Kahan, contains three specific disagreements over which the standard nearly foundered. The first was the format of subnormal numbers — the small, gradually-underflowing values below the smallest ordinary normalised number. Kahan's proposal (gradual underflow, in which subnormals occupy the space just above zero at the cost of a small amount of hardware complexity) was opposed by DEC's committee representatives (flush to zero, in which everything below the smallest normal number becomes zero, at a small cost in silent accuracy loss and no cost in complexity). The vote to include gradual underflow in the standard, at the p754 meeting of Wednesday, 22 September 1982, held at Larry Blake's, was eight to seven. Every commercial floating-point unit designed since — every one, without exception, from the Intel 80387 forward to the Apple M-series and the NVIDIA H100 — implements gradual underflow. The second disagreement was the representation of NaNs, and specifically whether the standard should require the propagation of signalling NaNs through arithmetic without user intervention. It should, Kahan argued, and it does. The third was the direction of default rounding. Kahan proposed round-to-nearest-even. DEC preferred round-to-nearest-away-from-zero, which is faster to implement in serial hardware and produces a small statistical bias in accumulated sums. Round-to-nearest-even carried by twelve to three at the meeting of Thursday, 3 March 1983.

Kahan received the Turing Award in June 1989 for his role in bringing floating-point arithmetic under the discipline of an interoperable numerical standard. The lecture, delivered at ACM Award Banquet in San Francisco and published in Communications of the ACM for June 1990, contains a paragraph which I quote without gasconade of my own (there is a callback in there, friend, for those following along):

"In the end, the standard is not what mathematicians would choose. It is what could be shipped in silicon and could survive the review of committee members who were, in the main, trying to protect their own installed base. That the standard that emerged is as good as it is — better, in most respects, than any of the private formats it replaced — is testimony to the willingness of the members, over eight years, to trade in their prejudices for something better."

He is, at the time of writing, ninety-three years old, an Emeritus at Berkeley, and, according to the departmental directory, still occasionally answering student email.

Primary sources:

IEEE Standards Association. IEEE Std 754-1985: IEEE Standard for Binary Floating-Point Arithmetic. Institute of Electrical and Electronics Engineers, New York, 12 March 1985. The formal standard: 18 pages including the recommended function definitions in Appendix A. Superseded by IEEE Std 754-2008 (revised August 2019) but essentially unchanged in its core representation and rounding rules.
Severance, Charles. "An Interview with the Old Man of Floating-Point: Reminiscences elicited from William Kahan." Interview conducted at Berkeley, 20 February 1998; edited transcript archived in the IEEE Annals of the History of Computing and available in full on Severance's own website. The single most complete source for the internal history of p754.
Kahan, W. "The Regrettable Failure of Automated Error Analysis." Turing Award Lecture, ACM Award Banquet, San Francisco, June 1989; published in Communications of the ACM, Vol. 33, No. 6, June 1990, pp. 76–82. Kahan's own retrospective, delivered forty-eight hours after the award ceremony.

Today's goal

Say, once today, out loud or in writing, a thing you like about your work. Not a thing your work has produced for someone else — that is a Grumbletonian's escape hatch, a way of praising your own utility rather than your work itself. A thing you like about the thing while you are doing it. The specific compile-error message that always makes you smile. The moment the SQL query planner returns a plan that is exactly the plan you had drawn on paper. The two seconds after a rebase when everything is quiet and green. Say the thing. Do not say it as a piece of therapy, or as a reframe, or as a mood management technique. Say it because it is true and you have not said it in a while. The Country Party would not approve. Say it anyway.

Today's toy is the ha-ha — a small interactive rendering of the eighteenth-century English landscape garden feature of the same name: an invisible ditch, sunk into a lawn and faced with stone on the pasture side, whose whole purpose is to keep livestock out of the ornamental grounds while producing no visible boundary from the manor's drawing-room window. Slide the observer's viewpoint across the lawn and watch the ha-ha do what a ha-ha does: nothing, from a distance, and then, at about twelve paces out, quite a lot, all at once. Charles Bridgeman and William Kent used them at Stowe from about 1719; the word is said, by Horace Walpole writing in 1780, to come from the exclamation you make when you nearly walk into one. Lives in the corner.

Go say the thing, friend.

— C