← Back to context

Comment by kragen

9 months ago

Crucial context here: Ousterhout is one of the great programmers who built the free software world we live in today, and Uncle Bob is a faker. Ousterhout is not without his problems (Stallman famously called him a "parasite" on the free software community, as well as fervently disagreeing with his technical taste) but he's written truly world-changing software. By contrast, Uncle Bob is a windbag book author who has never managed to write any software worth using, to my knowledge.

Ousterhout to Uncle Bob:

> maybe you were surprised that it is hard to understand, but I am not. Said another way, if you are unable to predict whether your code will be easy to understand, there are problems with your design methodology.

This debate is full of treasures like this. What a brilliantly clear and understated way to expose charlatanism!

What is this “world-changing software” I'm saying Ousterhout has shipped? Tcl. (Hold on, now, don't downvote just yet.) Tcl has been a crucial enabling technology for EDA and automated regression testing since literally the 01980s. Probably every VLSI chip in the computer you're reading this on was designed, verified, and tested with workflows involving unholy amounts of Tcl. GCC's test suite is also Tcl. Still.

Automated testing in the 01980s? Yes. It's true that automated testing wasn't very prevalent in the software world until the Agile guys (Uncle Bob and his less incompetent compatriots) popularized it around the turn of the century, but EEs and compiler engineers have been pervasively automating testing a lot longer than that, and Tcl was for a long time the least awful option, believe it or not. And that was John Ousterhout's doing.

Do you know what the SPICE developers did to make SPICE scriptable, before there was Tcl? They linked csh into it. Motherfucking csh. If you've never tried to maintain a large script in csh, you do not know the meaning of suffering.

Good programmers write good software; bad programmers write bad software, or no software. Ousterhout has written one of the few pieces of software that can be called great. (In its historical context. In 01978 csh was great software too.) What software has Uncle Bob written?

Listening to Uncle Bob's programming advice over Ousterhout's would be like listening to your middle-school English teacher's writing advice instead of Stephen King's. It's not that King could never give you worse advice, but if you need your English teacher's advice, generally your judgment will not be good enough to distinguish the rare occasions King gets it wrong.

6 comments

kragen

Reply
cmacleod4  9 months ago

I'ma big Tcl fan, but Ousterhout has created many other important things - see https://en.wikipedia.org/wiki/John_Ousterhout .

  • kragen  9 months ago

    I don't think Magic and Raft are anywhere close to the importance of Tcl, though I've probably at some point used a chip that was designed in Magic. And, while I like Tk and find it inspiring, the number of Tk apps I can remember ever using (that I didn't write myself) is maybe three, and they weren't very important to me.

    As for Sprite-LFS, I really enjoyed the Sprite LFS paper and found it inspiring, but my conclusion was that Seltzer's followup BSD-LFS paper falsified some of its more surprising claims, and ultimately the underlying predictions about the relative trends in RAM size and disk size turned out to be wrong, undercutting the key advantages of the LFS approach overall. Vaguely LFS-like approaches are important to SSDs and SMR disks, but WAFL was already about that LFS-like in 01995 (which is admittedly after Sprite-LFS), and SSD FTLs also do some not-very-LFS-like things. So ultimately I don't think Sprite-LFS turned out to be that important.

    Sprite as a whole I'm less able to evaluate. I've never been an OS researcher, but I've spent a fraction of my life reading SOSP and HotOS papers and systems dissertations, and I don't remember seeing anything that came out of Sprite except Sprite-LFS. I was thinking maybe doors in Solaris did, but no, that was Sun's Spring, not Sprite. Other side of the Bay, where Ousterhout took Tcl eventually. So it's possible Sprite was a great achievement, but I haven't noticed it. But I think more likely it's one of those things where we tried the "obvious" thing (SSI across a bunch of workstations) and found out why it was bad, which influenced later efforts like PVM, MOSIX, Beowulf, distcc, MapReduce, Ceph, etc., because Sprite stepped on the mines so they didn't have to. There's a nice retrospective (by Ousterhout, natch) at https://web.archive.org/web/20150225073211/http://www.eecs.b....

    So I don't think Tk, Magic, Raft, and Sprite-LFS really have the same level of significance as Tcl. Sprite maybe.

    I don't think it's bad to spend a lot of time and effort on things that turn out to not be very significant, for two reasons. One is that, after a long enough time, very little indeed remains very significant. (Who, today, can recount the disappointments of the Minoan queens?) The other is that things you could do that are significant—even for a little while—are usually things that will probably fail. So if you spend a lot of time doing things that might be significant, you'll fail at most of them.

    But in Ousterhout's case, one of those things did succeed brilliantly, and it was Tcl.