There were a bunch of other issues but, yes, the compiler was a big one from which a number of the other issues stemmed.

I’m curious what kind of code his 30 lines were - I’m betting something FP-heavy based on the public focus benchmarks gave thst over branchy business logic. I still remember getting the pitch that you had to buy Intel’s compilers to get decent performance. I worked at a software vendor and later a computational research lab, and both times that torpedoed any interest in buying hardware because it boiled down to paying a couple of times more upfront and hoping you could optimize at least the equivalent gain back … or just buy an off-the-shelf system which performed well now and do literally anything else with your life.

One really interesting related angle is the rise of open source software in business IT which was happening contemporaneously. X86 compatibility mattered so much back then because people had tons of code they couldn’t easily modify whereas later switches like Apple’s PPC-x86 or x86-ARM and Microsoft’s recent ARM attempts seem to be a lot smoother because almost everyone is relying on many of the same open source libraries and compilers. I think Itanium would still have struggled to realize much of its peak performance but at least you wouldn’t have had so many frictional costs simply getting code to run correctly.

This form versus substance issue is a really deeply embedded problem in our industry, and it is getting worse.

Time and again, I run into professionals who claim X, only to find out that the assertion was based only upon the flimsiest interpretation of what it took to accomplish the assertion. If I had to be less charitable, then I’d say fraudulent interpretations.

Promo Packet Princesses are especially prone to getting caught out doing this. And as the above story illustrates, you better catch and tear down these “interpretations” as the risks to the enterprise they are, well before they obtain visible executive sponsorship, or the political waters gets choppy.

IMHE, if you catch these in time, then estimate the risk along with a solution, it usually defuses them and “prices” their proposals more at a “market clearing rate” of the actual risk. They’re usually hoping to pass the hot potato to the poor suckers forced to handle sustaining work streams on their “brilliant vision” before anyone notices the emperor has no clothes.

I’d love to hear others’ experiences around this and how they defused the risk time bombs.

We're not living through this again at all with generative AI, right?

> “You're predicting the entire future of this architecture on 30 lines of hand generated code?"

It’s comforting to know that massively strategic decisions based on very little information that may not even be correct are made in other organizations and not just mine.