Carmack on Apple successfully migrating from 68k to PPC to x86 to ARM | Hacker News Reader

Carmack on Apple successfully migrating from 68k to PPC to x86 to ARM

twitter.com

54 pointstosh25 comments

A notable anecdote from Carmack's post:

"One time, my wife, then fiancée, and I were meeting with Steve at Apple, and he wanted me to do a keynote that happened to be scheduled on the same day as our wedding. With a big smile and full of charm, he suggested that we postpone it. We declined, but he kept pressing. Eventually my wife countered with a suggestion that if he really wanted “her” John so much, he should loan John Lassiter to her media company for a day of consulting. Steve went from full charm to ice cold really damn quick. I didn’t do that keynote."

Carmack's videos too are filled with fascinating technical details and thoughts. He rarely wastes words on fluff and is quick to point out shortcomings on his or his teams code as well as how to improve them. I find the way he communicates to be an inspiration on being a technical leader.

> I advocated repeatedly to Steve that an Apple phone could be really great. Every time there was a rumor that Apple might be working on a phone, I would refine the pitch to him. Once he called me at home on a Sunday (How did he even get my number?) to ask a question, and I enthused at length about the possibilities.

Imagine knowing in the era of J2ME that there will be the iPhone (and iPad) and phoning Carmack to get an 2 hour impromptu speech about mobile phones. I bet Jobs got a kick out of that.

First, I think the URL should be updated to https://raw.githubusercontent.com/ESWAT/john-carmack-plan-ar... and skip all the steps of clicking through multiple tweets.

Second, the title is inaccurate at best. The original content does not seem to have a title, but in the spirit of the OP, something like "Carmack on Apple migrating from 68k to PPC" would be fitting.

Third, I would suggest we put the year (1999) in the title.

> Apple successfully migrating from 68k to PPC to x86 and now on to ARM is quite the journey

I found the tweet itself noteworthy, in it Carmack is also referencing his own facebook post on Steve Jobs for context

https://www.facebook.com/permalink.php?story_fbid=2146412825...

previous discussion: https://news.ycombinator.com/item?id=17066846

ya some date add please, at least (2018) or (1999)

Pastebin, because the rest of the content is behind facebook:

https://pastebin.com/gdq56jBQ

Apparently the Facebook story is not the content that was intended to be shared, but rather a story linked in a tweet that is linked in the tweet that was shared, below the Facebook story that was linked:

https://raw.githubusercontent.com/ESWAT/john-carmack-plan-ar...

This appears to be an article from 1999, and does not talk about the more recent migrations between instruction sets.

Cases like this suggest that a serious portion of the HN users just upvote posts based on the title without ever opening them. The current title is very misleading.

This tweet share on HN may need some explaining or instructions. It appears to be highlighting a tweet that links to a Facebook story, but reading through that story, I don't really get any sense of Apple's migration through CPU instruction sets. Glancing back at the tweet, it might have a thread of comments where the M1 is discussed, but I can't really follow it, so I'm not sure.

The tweet also links to another tweet [0], which links to a GitHub file [1] with the story referenced in the title.

[0]: https://twitter.com/brockgs/status/1330711644977053696

[1]: https://raw.githubusercontent.com/ESWAT/john-carmack-plan-ar...

Thank you. The story appears to be from January 10, 1999 and does not discuss Apple successfully migrating from PPC to x86 to ARM.

btw the text file is from the John Carmack plan archive:

https://github.com/ESWAT/john-carmack-plan-archive

From linked FB post:

> Steve first talked about application development for iPhone at the same keynote I was demonstrating the new ID Tech 5 rendering engine on Mac, so I was in the front row. When he started going on about “Web Apps”, I was (reasonably quietly) going “Booo!!!”.

I think Carmack was not alone in that assessment. A year later Apple delivered not just what developers asked for - an SDK - but an SDK and the App Store.

Turns out Jobs was kind of right about web apps: not only did Safari support user interface elements and multitouch out of the gate, but years later in the age of Webasm + Canvas + WebGL we are at the point where Luna game streaming is rolled out as a web app in Safari.

Are there any 68k apps that you can run on ARM MacOS though?

Could you even recompile them?

Not really. If you really want to, you can probably compile about everything, as hat just needs header files that declare what functions used to exist.

Those are easily copied over from old systems, once you have found them. You may have to redefine some basic types, but that’s all doable.

However, I can’t think of _any_ function defined in a library on modern 64-bit Mac OS that even has the same name as that of one in classic Mac OS, so linking would be impossible.

afaiu this doesn't go back as far but it spans back to Classic Mac OS:

https://www.highcaffeinecontent.com/blog/20150124-MPW,-Carbo...

I particularly don't think this is remarkable (from the software side, logistics is another thing). Most UNIX software has been portable between CPUs since at least the mid-70s. Apple is doing what UNIX pioneered 40 years ago.

Sort-of. You can compile it if you have the source. Commercial software (binaries) are often unable to be run outside of their target cpu and flavor of Unix though. Apple is a little different in that FAT binaries can contain multiple architectures and even binaries that are not built for the current arch are automatically run in an emulated environment.

Solaris had a mechanism where one could place binaries in arch-specific folders (eg: /bin/sparcv9) and even per-arch library paths but it was definitely opt-in with varying levels of adoption.

I’ve also seen some pretty fun hacks to start qemu in Linux for certain non-native executables. Still not nearly as complete/smooth/user-friendly as the fat binary/library approach.

Right. Because 40 year old UNIX is equivalent to the behemoth a Full-fledged 2020 desktop/Notebook operating system.

There’s probably millions of bits and bytes that are not trivially portable and more for a single person to even grasp an overview of in his or her lifetime. There were assumedly tens of thousands of software engineers involved.

But „it’s not remarkable“. Yeah.

Please don't take threads further into flamewar or break the site guidelines by being snarky. You have a good point here but it's important to make your substantive points thoughtfully.

https://news.ycombinator.com/newsguidelines.html

Not really. For another example, look at Debian Linux. They have several million lines of code supported on multiple architectures. And Apple didn't do this from one day to the next, they have the OS running on ARM CPUs in one version or another for more than 10 years.

So it’s not an impressive amount of work because it has taken them 10 years of effort?

No, because it has been done before by much less powerful institutions.

UNIX did not pioneer what other mainframes did since 1961, just some UNIX fans think it did.

Luckily there are papers and documentation from the work going outside Bell Labs.

Reply on news.ycombinator.com