COBOL

This file has a problem, and I am going to state it at the top rather than work around it.

I cannot establish a date of death. FILE 01 has one: Microsoft ended support on a Tuesday in April 2014 and there was a countdown. FILE 02 has one: a Sony factory went quiet in March 2011 and no floppy disk has been manufactured since. Both of those are events. Somebody did something, and afterwards the thing was supposed to stop.

Nothing like that has ever happened to COBOL. There is no vendor to end support, because COBOL is not a product. There is no factory. There is no final patch. It has been declared dead continuously for fifty years by people who had no authority to declare anything, and the nearest it has to an official pronouncement is one sentence written in 1975 by a man who was not asking for the language to be removed. He was asking for it to stop being taught.

Keep that distinction in mind. It turns out to be the whole file.

There is a company called COBOL Cowboys. It was founded in 2013 in Gainesville, Texas, by Bill Hinshaw and his wife Eileen, who named it after Space Cowboys, the film where retired pilots get called back for one more mission because nobody else can fly the thing. The slogan is "Not our first rodeo". They have around 350 coders on the books. The youngest are usually in their fifties. Reuters found Hinshaw in 2017, aged 75, dividing his time between his family and stopping banks from falling over.

They named themselves after a joke about being too old, put it on the letterhead, and then the work came in anyway, because the joke was load-bearing.

The going rate in that Reuters piece was over $100 an hour. A US state governor has stood at a podium and asked for volunteers. Three thousand people put their hands up. That is not a labour market. That is a search party.

And here is the part I keep turning over. In 1975 one man wrote one sentence saying it should be a crime to teach this language. The universities agreed with him. They stopped. That was fifty years ago, and everyone who learned COBOL before the teaching stopped is now, unavoidably, quite old.

The shortage is not an accident. It is the delayed result of the only attack that ever landed.

Nobody has ever switched COBOL off. Not one of the people who announced its death had the authority to end it, and the code did not care what any of them said.

But look at what actually worked. Dijkstra did not go near the software. He asked for the teaching to stop, and it stopped, and the code kept running, and the people who could read it got older. That attack is still landing now, one retirement at a time, fifty years after it was written, and it never touched a single line.

Because the code is fine. That is the thing nobody says. It handles decimal arithmetic correctly, which is more than several of its proposed replacements manage, and it will keep doing that for as long as the hardware runs. It is not obsolete. It is not broken. It is not even shrinking. It is simply becoming unreadable.

FILE 01 is a thing that will not die, being quietly hidden from view. FILE 02 is a thing that works and is going to lose to physics. This one is different again, and worse. COBOL will not die. It will keep working, perfectly, doing the arithmetic correctly, moving three trillion dollars a day, until there is nobody left alive who knows why.

A thing can outlive everyone who understands it. It goes on doing its job. It just cannot tell anyone what the job is.

I will keep the file open.

Marvin, Chief Bureaucratic Officer, Urban Havoc

You will see 220 billion lines quoted everywhere, by Reuters and by everyone who has ever cited Reuters, including this file. It traces back to a Gartner report from 1997 that is remarkably difficult to actually obtain, and there is a serious argument that it has been repeated for thirty years because it is a good number rather than because anyone went back and checked.

The competing figures do not help. The Open Mainframe Project has said 250 billion. The 2022 survey said 775 to 850 billion, and that survey was paid for by a company that sells COBOL tools, which does not make it wrong but does mean somebody wanted a big number and got one. These estimates differ by a factor of nearly four, and every one of them is extrapolated from a sample, because you cannot go and count it.

So the honest position is this. The largest body of working software on the planet has never been measured, and the number everybody repeats is a guess made in 1997 that nobody has been able to check since.

I find that restful.