[awakened]

Many years ago, I submitted a bug report to OpenBSD about an issue I had discovered with threads. I received a one line response from Theo. I still have the mail. He wrote:

   "Threads are for idiots."

At the time, I felt discounted and I was upset. I was younger then. Today, I realize what he meant and that he's right.

[tptacek]

This is vintage Theo. It has the virtue of sounding correct, but, because OpenBSD ships threaded libraries, it probably lacks the virtue of being correct.

I had a similar experience (note, though: I have a history with Theo, who I know/knew personally).

When I was at Arbor Networks, we shipped appliances that monitored ISP backbones that were based on OpenBSD. An analysis process that happened to allocate a lot of memory would occasionally lose a giant chunk of memory. I was able to produce a reduction of the bug and narrow down where in the VM subsystem the bug was happening, but I wasn't able to recommend a fix. Theo's response, to what was clearly a serious bug in OpenBSD, was "I'm not going to look at UVM; it's just Chuck Cranor's thesis project".

I lobbied for a switch to FreeBSD, but the monkey.org people that ran the place were dyed-in-the-wool for OpenBSD. :)

[Flow]

What happened next? Did the bug get fixed? Did you do a work-around somehow? Don't leave stories unfinished like this :)

[tptacek]

I actually don't know. Arbor could have had its dev team hunt for a fix for the bug, but that would have been silly; no way did it make sense for them to take ownership of a custom fork of the most complicated kernel subsystem. So we worked around the problem instead.

[pvg]

In the beginning, when Theo was cast out of NetBSD, someone rewrote this thing

http://www.skrause.org/humor/poohgoesapeshit.shtml

with Theo and other BSD-ish personalities as the characters. There are many stories of prickly people mellowing with age. It's unlikely there'll be such a story about him.

[gypsyharlot]

I don't see how his response was unreasonable.

[tptacek]

It helps to understand the role a virtual memory system has in an operating system kernel.

[fabulist]

Its also important to choose your security battles. I imagine there was another severe bug or two in a project as large as OpenBSD.

That being said, thats a petty and ridiculous reason not to deal with someone's code. This gentleman strikes be as a builder, not a maintainer; someone who wants to breeze through town like a cowboy, bring cool ideas to fruition, and move on to the next conquest while someone else keeps the system from falling apart. Thats not a judgment, we need people like that, but they probably shouldn't be the ones fielding emails.

[markild]

> I realize what he meant and that he's right.

Could you please expand on this?

[bananas]

I think the rise of async/select based and privilege separated designs tend to suggest that threads don't work well for system programming.

[tptacek]

This is an incoherent response. Privilege separation is orthogonal to async designs. It's just as easy to privsep a synchronous program. Meanwhile, while I happen to appreciate async designs, it's far from settled as to whether they're long-term sounder than thread. What I know from experience is that it's easier to make async designs performant. Nothing I've seen suggests that they're that much safer.

[bananas]

Sorry I agree it was rather incoherent - I was distracted half way through posting.

I'm not suggesting they are sounder but I'm suggesting that async designs are simpler and simplicity rules when it comes to safety. sync designs tend to evolve into complexity over time to maintain performance (IIS for example which is a behemoth of threaded privsep pain).

And in my experience (so totally an anecdote and I accept that), it's not easier to privsep a synchronous program. On top of the IPC concerns of isolation, you still have all the problems associated with threading. It's just pain.

(I've written a fair number of both types of systems - none open source unfortunately)

[tptacek]

The original privsep programs weren't async reactor-based designs.

[bananas]

No they were single threaded forked. Aware of that.

[yp_all]

Great story. It is good to know that some programmers do learn over time.

For every 1 story like yours there are probably 10 where someone is still traying to get "revenge" by saying nasty things about you know who.

Those more common stories get replicated through the wires and the end result is that no one cares about programming anymore; they are simply interested in discussing interpersonal relations.

Would you rather have your software written by a foolish programmer who is pleasantly tactful or a competent and conservative one who is tactless?

The answer of course varies depending on what you are really after.

[JackC]

> Would you rather have your software written by a foolish programmer who is pleasantly tactful or a competent and conservative one who is tactless?

This is a false dichotomy. Interpersonal skills are among the skills necessary to work on a project that involves more than one person. That covers most of the interesting ones. If there's a successful project with someone who lacks those skills, it's because other people are covering for them -- just like a successful project with a foolish programmer. Sometimes that person makes up for it with other things they bring to the table, and that's fine, but don't pretend a weakness is a virtue.