Hacker News Reader

Comment by tenebrisalietum | Hacker News Reader

Having few instruction set architectures, or even one, enables quicker and more efficient software development by lowering barrier to entry (easier to learn and study just one ISA instead of many) and lowering the amount of maintenance and development work that has to be done on development infrastructure such as compilers.

This is assuming royalties don't have to be paid to an ISA creator, which is a separate issue.

A single ISA doesn't have to mean there is a single CPU vendor or CPU manufacturer.

View on HN

In my opinion, having a single ISA breeds lazyness, and causes shortcuts to be made in programming. Like when the whole world got rid of all those esoteric 18-bit-word and worse architectures, and started writing loads of C for 32-bit i386. Then 64-bit came along and loads of software wouldn't compile properly without fixing things like assumptions of integer size, etc.

It's why I am very glad that while Linux is very popular, there is still a BSD keeping programmers from going too crazy on the Linux-only features.

We don't know what future architecture will exist. I would rather we have a multitude of architectures now, so that people write standards-conforming code rather than it-works-on-amd64/Linux code.

The 18- ad 36-bit architectures weren't esoteric at the time! Though may of them also had bad (outdated, and not-obviously bad at the time) ideas like one's complement arithmetic. Machines like the PDP-10 were a pleasure to program in assembly, but I haven't written a line of assembly code in over a decade.

But I agree the world is full of "slap it together and if it runs once, ship it" software -- that's what gave us Y2K. Time to re-read worse is better.

In retrospect, it's amazing how older games like Age of Empires work so well and (almost) bug free from the get go. In the software development environment for late 90s the amount of work to making sure that it works (without crashing!) is amazing.

With extreme logging now our team still throws a bunch of internal errors today, granted our team is much smaller but still, those older games are amazing.

"If you're not working with 36 bits, you're not working with a full DEC!" -DIGEX (Doug Humphrey) trolling the 32-bit VAX weenies at DECUS.

https://news.ycombinator.com/item?id=13514918

> In my opinion, having a single ISA breeds lazyness, and causes shortcuts to be made in programming.

Aren't you aware that lazyness is one of the great virtues of a programmer?

> Then 64-bit came along and loads of software wouldn't compile properly without fixing things like assumptions of integer size, etc.

Yes, now you just write code with the assumption that a word is 4 or 8 bytes, you don't concern yourself with the case where the word is four triganic nibbles long.

OpenBSD folks think supporting more architectures have helped them creating a better OS.

Yup...Theo specifically talks about alignment restrictions on SPARC as catching all sorts of weird corner-case bugs.

I think it is appropriate to specify your statement with a significant detail; a ISA monoculture enables quicker and more efficient development of proprietary software. Free and open software deals efficiently and fast with different ISA's, as can be seen in the ports of Debian and the targets of GCC.

Free software is great, but it's not efficiently developed. That's why it's been so hard for people to get a satisfactory Linux on the desktop GUI and it's still a crapshoot to even get all of a laptop working .

> Having few instruction set architectures, or even one, enables quicker and more efficient software development by lowering barrier to entry

> A single ISA doesn't have to mean there is a single CPU vendor or CPU manufacturer.

Then let's standardize on a single open and royalty-free one. That excludes x86 and ARM.

I agree that there's a programmer cost to having a multiplicity of ISAs. However, ISAs still have some relation to the design of the microarchitecture, which for x86 processors is not open. The RISC-V ISA for instance was created mainly because it helped the effort to create open CPU hardware.

My usual take is that it gives us better efficiency in the short-term, at the expense of making it far more likely that we get stuck in a rut long-term.

Kinda like Japan's economy. Lots of old companies trying to keep up and very little outside entrepreneurship to disrupt and progress the industries.

Reply on news.ycombinator.com