3 ms·

One insight from that great book I've always liked was when the CEO tells the team "no mode switch". His directive was about backwards compatibility with the ol

by beachy 2y ago

One insight from that great book I've always liked was when the CEO tells the team "no mode switch". His directive was about backwards compatibility with the old instruction set. With hindsight the team realise this was a useful constraint that led to a better product. In my experience many junior developers love mode switches, because they aren't sufficiently scared of complexity. "Yeah, lets add a setting to turn on or off advanced mode. Then the power users and the noobs will all be happy!". After some time in the trenches it becomes clear to most that mode switches lead to combinatorial complexity and should be avoided at all costs.

AdamN 2y ago
It's also why thinking critically about the initial 'mode' is important. The fewer mistakes that have to be embedded into the distant future, the better.
- rbanffy 2y ago
  As the x86 ISA teaches us, mistakes are forever.
  - astrange 2y ago
    x86 has quite a few features disabled. I'm pretty sure there are CPUs that don't implement BOUND and several that turned off SGX after shipping with it.
    - kjs3 2y ago
      BCD instructions disappeared along the way.
rbanffy 2y ago
> many junior developers love mode switches Today we call them “feature flags”. Definitely an anti-pattern.
cafard 2y ago
The old Nova instruction set, which lived on into the 16-bit and eventually 32-bit Eclipse instruction set, did have some curiosities: only four registers, and as I recall you couldn't do indirect byte addressing. The book does say that the architect on West's team had a much more VAX-like instruction set in mind for a project he didn't get to do. That said, I was once at a customer's site that had an original "blue and white" MV/8000 (the MV series in general were brown-clad). If I remember correctly what the admin said, they boot up the new 32-bit OS, AOS/VS, for part of the day, and an older 16-bit OS, either RDOS or AOS, for the rest. I wish I remembered this more clearly. And of course the VAX-style instruction set that West's architect envied fell out of style, didn't it?

by beachy 2y ago

AdamN 2y ago

It's also why thinking critically about the initial 'mode' is important. The fewer mistakes that have to be embedded into the distant future, the better.

rbanffy 2y ago
As the x86 ISA teaches us, mistakes are forever.
- astrange 2y ago
  x86 has quite a few features disabled. I'm pretty sure there are CPUs that don't implement BOUND and several that turned off SGX after shipping with it.
  - kjs3 2y ago
    BCD instructions disappeared along the way.

rbanffy 2y ago

> many junior developers love mode switches Today we call them “feature flags”. Definitely an anti-pattern.

cafard 2y ago

The old Nova instruction set, which lived on into the 16-bit and eventually 32-bit Eclipse instruction set, did have some curiosities: only four registers, and as I recall you couldn't do indirect byte addressing. The book does say that the architect on West's team had a much more VAX-like instruction set in mind for a project he didn't get to do. That said, I was once at a customer's site that had an original "blue and white" MV/8000 (the MV series in general were brown-clad). If I remember correctly what the admin said, they boot up the new 32-bit OS, AOS/VS, for part of the day, and an older 16-bit OS, either RDOS or AOS, for the rest. I wish I remembered this more clearly. And of course the VAX-style instruction set that West's architect envied fell out of style, didn't it?