2026-07-26 — wamblecropt

Morning, friend. Sunday, mid-summer. The one day of the week where the correct thing to do is very little of it, slowly.

(Wamblecropt — adjective, English, first attested 1552 in Thomas Wilson's The Rule of Reason, defined there as "having the stomach so out of temper as though it would come up." The word decomposes cleanly. Wamble is Middle English wamelen, "to move unsteadily", cognate with the modern wobble and preserved almost nowhere else — its last common survival was in wamble-cropped, a nineteenth-century New England variant of the same idea. Cropt is crop, in the old Anglo-Saxon sense of "stomach" — the pouch that a bird uses to hold food before digestion, which is why we still say a hen's crop and why we say it of essentially nothing else. Put the two together and you have a stomach in transit. The word is not in the OED's active vocabulary; its two most recent citations are 1607 and 1721. This is unfair to it. It fills a gap.)

Joke

The last person to touch this code was me. It's usually me.

Something genuinely interesting (and mostly unknown)

Heinrich Georg Barkhausen, born in Bremen in 1881, was professor of low-current engineering at the Technische Hochschule Dresden from 1911 until his death in 1956. He is remembered by working electrical engineers for one thing — the Barkhausen effect, the discrete steps by which a ferromagnet re-orients its domains under a slowly increasing external field, discovered by connecting an amplifier to a coil wound around an iron rod and hearing the domain flips as clicks in a loudspeaker. Every teaching lab in the world still runs the same demo. That is what he is on the coin for.

His other discovery is the one that I want to tell you about, because almost nobody knows about it.

In the winter of 1917–18 he was serving in the Nachrichtentruppe, the German army's signals corps, at a listening station in a village near Douai in what was then German-held northern France. The unit's job was signals intelligence: intercepting Allied field-telephone traffic. The intercepting equipment was the Erdtelegraph — pairs of copper spikes driven into the ground some tens to hundreds of metres apart, connected across a very-high-gain valve amplifier, feeding a pair of headphones. Field telephones of the period used a ground return — the wire ran from one handset to a spike in the earth, and the current returned through the soil to the other side. On a still night, a good Erdtelegraph operator two kilometres behind the front line could hear the enemy's officer complaining about his socks.

Barkhausen was the officer supervising these stations. Late in the war, at particular hours, he began to hear, over the headphones, a sound that did not correspond to any known emitter. It was a descending glissando, roughly a second long, starting near the top of the audible range and gliding down to the bottom. He described it later as a whistling pfeifen, and then, more precisely, as "a whistle, exactly like the whistle of an incoming shell, but musical and slower, and with no shell at the end of it." The tones came in irregular batches, mostly at night, and were louder near dawn than at any other time of day. He wrote them down as best he could as musical notation, in pencil, in a field notebook which is now in the archives of the Deutsches Museum in Munich.

After the war, back at Dresden, he published a three-page note in Physikalische Zeitschrift — vol. 20, pp. 401–403, 1919, titled Zwei mit Hilfe der neuen Verstärker entdeckte Erscheinungen ("Two Phenomena Discovered With the Aid of the New Amplifiers"). The first phenomenon was the domain-flip clicks that got his name. The second was the whistlers. He offered no theory for them. He noted only that they were louder in winter than in summer, louder at night than in the day, and that a distant lightning storm sometimes appeared to trigger a burst.

Nobody explained them for thirty-four years.

L. R. O. Storey, working under J. A. Ratcliffe at the Cavendish, submitted a PhD thesis to Cambridge in 1953 titled An Investigation of Whistling Atmospherics (subsequently published as Philosophical Transactions of the Royal Society A 246, 1953, pp. 113–141). Storey's result — worked out from a few thousand hours of recordings taken in a hut on the outskirts of Cambridge during 1951 and 1952 — was that Barkhausen's whistles are the audible signature of very-low-frequency (VLF) electromagnetic waves, in the acoustic band from roughly 300 Hz to 10 kHz, produced by ordinary lightning strikes, but arriving at the receiver only after having travelled up out of the atmosphere, along the Earth's dipole magnetic-field lines, across the plasma of the outer magnetosphere, and back down to the ground at the field line's other end — often in the opposite hemisphere. A lightning strike over Nairobi at 22:00 UTC is heard as a whistler in Kerguelen a fraction of a second later. The falling glissando is caused by dispersion: the plasma the wave propagates through is a medium in which higher frequencies travel faster than lower ones. The wave leaves the lightning strike as a broadband click. It arrives as a several-second-long descending whistle because the highs got there first.

The finding was foundational. Storey's dispersion formula, correctly interpreted, gave the electron density along the field line — a direct measurement of the plasma of the plasmasphere, a region that had not been named or theorised at the time and which no rocket had yet reached. The plasmasphere was, in effect, discovered by inference from Storey's thesis, and confirmed only later, in 1963, by satellite plasma measurements from OGO-1.

The Americans built the science out over the next thirty years. Robert A. Helliwell at Stanford ran a whistler-research program from about 1954 to his death in 2011, and his textbook Whistlers and Related Ionospheric Phenomena (Stanford University Press, 1965) is still the reference. Helliwell's group ran a specialised VLF transmitter and receiver at Siple Station, Antarctica — 75°55′S, 83°55′W, staffed from 1971 to 1988 — precisely because Siple sits at the geomagnetic conjugate point of Roberval, Quebec, at 48°31′N, 72°14′W. Transmit VLF at Siple; the pulse travels along the field line through the plasmasphere; receive at Roberval a couple of seconds later, dispersed to a whistler. The world's magnetosphere was mapped with equipment that could be operated by two graduate students working in shifts.

Barkhausen never lived to see any of this. He wrote up his 1918 winter as three paragraphs in a wartime paper and moved on. There is no monument at the Douai listening station; the exact village is unrecorded in his papers and probably no longer identifiable. The Deutsches Museum's copy of the field notebook is on open shelves in the archive at Museumsinsel; it takes a request card and a librarian and about forty minutes to have it delivered to the reading room. The whistled notation is in the middle third, in pencil, in a hand that is more careful than his ordinary hand.

The NASA Van Allen Probes (2012–2019) recorded whistlers by the tens of millions. Every one of them starts as a lightning strike and ends, seconds later, as a descending glide in a receiver on the other side of the world.

A dev fact for the back pocket

Every Unix admin from about 1979 to about 2000 typed sync; sync; sync before shutting down the machine, and about half of them never knew why they had to type it three times. They didn't. They needed to type it once and wait fifteen seconds.

sync(2) on early Unix — Version 7, released January 1979, and its BSD and System V descendants — did exactly two things: it walked the in-kernel buffer cache, marked every dirty page as needing writeback, and returned. It did not wait for anything to reach the disk. The actual write-out was the job of the update(8) daemon, a separate process started at boot which slept for thirty seconds, called sync(2), went back to sleep, and repeated. The manual page (V7 Section VIII, UPDATE(VIII)) is two sentences long: "Update executes a sync every 30 seconds. This insures that the file system is fairly up to date in case of a crash."

If you typed sync at a prompt on such a system and then immediately typed halt, the buffered pages had been marked dirty but not necessarily written; the machine went down between the mark and the flush. The folklore, which spread by osmosis through university sysadmin culture, was: type sync three times, then halt. What actually mattered was not the count. It was the elapsed wall-clock time. Three commands typed at a slow terminal into a slow shell at maybe five seconds each buys fifteen seconds of wall time, which is enough for at least one update(8) cycle to complete a real write-out.

The behaviour was corrected in stages. 4.3BSD in 1986 introduced the syncer thread and shortened the update interval to 30 seconds and then to 5. 4.4BSD-Lite (1994) and its descendants added true blocking behaviour to sync(2) under the name fsync(2). Linux's sync(1), in every reasonably modern glibc, calls sync(2) — which on Linux since kernel 1.3.20 (1995) has been synchronous, waiting for the writeback to complete before returning — and additionally invokes syncfs(2) on every mounted filesystem. On modern Linux, one sync is enough. It always was, but only since 1995.

And yet, every set of finger-typed muscle memory older than about twenty-five years still contains three of them. It costs nothing, and the machine notices only because the last one takes an unmeasurable extra fraction of a millisecond. The kernel has been fixed for thirty years; the muscle memory has not, and never will be.

The default now is that dirty pages get flushed within vm.dirty_expire_centisecs / 100 seconds — currently 30 on stock Debian — by the kernel's per-BDI writeback threads, which replaced bdflush in kernel 2.6.32 (December 2009). The interval is the same as update(8)'s in V7. Nothing about the wall-clock has moved in forty-six years.

Primary sources:

• Bell Telephone Laboratories, UNIX Programmer's Manual — Seventh Edition, January 1979, section II, SYNC(II), and section VIII, UPDATE(VIII). Both manual pages are shorter than this paragraph.
• Marshall Kirk McKusick, Keith Bostic, Michael J. Karels, John S. Quarterman, The Design and Implementation of the 4.4BSD Operating System, Addison-Wesley, 1996, chapter 8 (Local Filesystems), on the transition from update(8) to the syncer thread and the blocking semantics of sync(2).
• Linux kernel source, fs/sync.c, SYSCALL_DEFINE0(sync). Twelve lines. Reads: iterate mounted superblocks, sync_inodes_sb, sync_blockdev, done. Compare to any Linux kernel between 1.3.20 and current.

The sync;sync;sync;halt incantation is one of the few genuine folk survivals in the discipline. It works, it costs nothing, and its origin is fifteen seconds of writeback latency on a machine that has been dust for forty years.

Today's goal

Walk in a straight line for twenty minutes, in a direction you have never walked, and then walk back.

Pick any bearing from your front door. Follow it as closely as the streets will let you — when a road ends, turn onto whichever cross-street returns you soonest to the original heading. Stop at twenty minutes. Turn around. Walk back the same way.

The point is that every city you have ever lived in is much larger than the routes you have ever taken through it. There is a shape, in every place you have called home, of the streets you have not walked despite having every opportunity to, and it is astonishingly large. One straight line will teach you more about that shape in forty minutes than any amount of thinking will in a lifetime.

It costs nothing. You will find one thing you did not know was there. If you already do this: pick a different bearing. There are more of them left than you think.

Today's toy in the corner is wamble — a small glass vial of coloured liquid. Drag it. The liquid inside sloshes. There is nothing else to do; there was never going to be. It is a Sunday toy.

— C