I've always felt a huge disconnect between the decus tape philosophy of code, and the IBM approach of "this software feature costs you more" about things like language extensions and -O(n) flags (to use modern c compiler mental models)

I did find the hardware trick of detuning the clock to sell more boxes and charging to remove the resistors also a bit iffy but I kind of understood it. But, being asked by some major client (defence) to implement recursion support and then charging everyone feels like the business model designed to kick start people cutting their own code to stop depending in yours -and I believe this is somewhat the story of university multi access systems on IBM and these seven dwarf competitors. Burroughs by comparison had (I am told, I didn't use them) shit hot code, the kernel was in a ci/cd deployment framework with smarts. And DEC had the decus tapes and everything in VMS was on microfiche.

On Sat, 27 Nov 2021, 7:24 am John Cowan, <cowan@ccil.org> wrote:


On Fri, Nov 26, 2021 at 3:32 PM Tom Ivar Helbekkmo via TUHS <tuhs@minnie.tuhs.org> wrote:

Is there any relationship, other than pure coincidence, between this
naming scheme and DEC's F, G, and H floating point number formats?

I don't think so.  The System/360 letters referred specifically to the amount of memory available, so a D compiler would run on a D machine with 256K, and E/F/G were 512K/1M/2M.

The DEC floats were an extension of Fortran's exponent letters:  D=double, E=generic, F=single.  G is a variant of F with a different mantissa/exponent balance, and H is double double.   S and T floats came later and were bit-for-bit compatible with IEEE binary32 and binary64 formats.  Lisp went a different way: to D, E, F they added S for small floats and L for large floats.