Bram Cohen

You're wrong about Apache having no responsibility here. There's a serious problem, causing a lot of real pain, and they can trivially fix the problem, with zero ill effects. To not do the right thing under these circumstances is simply being an asshole.

Even if they then decided to maintain the de facto mimetypes standard (which they wouldn't have to - BitTorrent is far more successfull than everything else), it wouldn't be much work at all, and in fact it's already been done. This beurocracy is a comical example of Bram's law in action.

(a) isn't going to happen because it would cause a lot of completely unnecessary pain. (b) isn't going to happen unless the ietf changes their firm policy against declaring application/x-* a no man's land which they make no truck with, and getting them to show some practicality there seems even less likely than getting Apache to realize that it's okay to do things in the section of the standard which the standards committee has abdicated opinion on.

I think that the person that is trying to put some sanity in the httpd codebase for OpenBSD is "Daniel Ouellet"

The purpose of httplib is to be a library for http, not an experiment in 1.1 functionality. As a library, it is unacceptably unusable, and that is the author's fault. The rest of Python's standard libraries is designed to be all you need, and generally succeeds, because those authors actually care about the quality of their work.

I've spent a fair amount of time looking into various Python http libraries, as have several other people, and we have yet to find one which one can simply just use. In particular, anything which requires blocking calls is simply unacceptable, and most of them are written that way. The twisted one seems like it should be promising but I can't figure out how to cut out just the http functionality to use elsewhere.

Of course I filed a bug report, and supplied a stack trace. There isn't a simple repro because it only happens flakily. My best guess is that it depends on how many bytes get read in at a time over the wire. Several people have tried to dig into the code and figure out what's going on, and have gotten nowhere. It's an un-debuggable mess.

I'm on your (BC) side in the tribal sense versus the Apache team. I mean when I work with some swiss army knife software product, half the problem is breaking through what the documentation thinks you want to know. For example, using gcj it took me 10 minutes to find out that I had to include --main to let it know which file has main(). I had to wait until I broke down and made a -2 level of stupid query on Google instead of assuming the online documentation would help.

I once told my little brother that the majority of working [with Windows] was to trick it into doing what you wanted (consider multiple "mastermind" configurings of device drivers). Software has matured but the tricking yourself into doing the imperfect to control perfect code is always going to be there. And it grates since all the code is perfect. That's got to be a missing part of the list you made -- learning to be a programmer -- because nothing gets built right. If software were built right then it would be VMS and only be programmed by chess grandmasters.

In a non-tribal sense, it's worth considering that once Apache refused 1/3 of the web browsers on the Internet since AOL was being HTTP incompatible. (disclaimer: that number seems impossible, shrug.) Also part of the realpolitick protocol of interfacing with others as an owner of a system component, is that you can't be as generous as human nature provides or it's a bofh scenario. So some of the cold shoulder is part of keeping not the Apache dev team sane, but their software sane.

p.s. I enjoy your http://bitconjurer.org/ page.
disclaimer: my_comment_post_value * my_coding_contributions = low_number.