Except
> "It turns out, there was a bug in Set.intersection(_:), but it had only been discovered this past June, and the fix hasn’t made it into a public version of Swift yet. "
One of the guys in my team handled a frantic support request from a Formula 1 team because they required reproducible builds or else they got some fine from the FIA, and they somehow hit a situation where our compiler was outputting different code depending on time of day. Wild.
The compiler itself felt well-written and there were some very smart people involved in developing it. And now that I think about it the bug wasn't different code being output but some value elsewhere in the generated binary itself - maybe some ELF header? I wish I could remember
I thought you meant TI. Their compiler would compile in the date of compilation as a string by default (it wasn't deadcode eliminated in the linker because it was sucked in their panic libraries, if memory serves).
In any case, yeah, ADI.
I used to have a methodical way of investigating most bugs: (1) here's what I know, (2) here's what I don't, (3) here's the thing I would like to know to most restrict the remaining search space. It was fine enough for awhile, but when you don't know things like "I wrote a 0 here, therefore a 0 exists here", or "my code explicitly uses aligned 32-bit reads, so I issued an aligned 32-bit read here", the errors in your assumptions quickly get amplified into nonsense. Nowadays I explicitly includle miscompilation (interpretation), bit-flips, and other such garbage as I write down what I do or don't know about the system, and I throw in ballpark probabilities to figure out where to look next. Most debugging isn't noticably slower that way, but the hairy problems are easily 10x easier when you haven't conditioned yourself to ignore the real solution.
It turns out the bug fix has been released publicly, but only on newer OS versions.
I ran into the same bug, but since I'd included a `precondition` to verify that each elf-group had only one common item type, I discovered the bug immediately. I then needed about 5-10 minutes to convince myself that it really was a Swift bug and not a mayoff bug.