Mailman 3 May 2002 - Twisted

[Twisted-Python] win32 event loop
by Itamar Shtull-Trauring 08 May '02

08 May '02

The problem - WaitForMultipleObjects, which is like poll() for Windows, won't work with more than 64 events. MS did this since apparently it scales really really badly. [Subtext: Microsoft hates me] So I've done some research, and it looks like I/O Completion ports are the way to go for scalable networking. This will probably mean rewriting (or subclassing) the stuff in tcp.py and abstract.py, since this the Proactor pattern, not the Reactor pattern, so things work a bit differently. I'm not even sure how UDP is done, but that's something I'll worry about later. [Subtext: Microsoft create more useless work for me] Another issue is the SSL support, which won't work with this method... We'd have to use Windows specific SSL code. [Subtext: Microsoft breaks my code] Another alternative is using threads with multiple WaitForMultipleObjects. This, I would guess, is probably really slow. [Subtext: Microsoft slows down my software]

3 3

[Twisted-Python] failure.Failure
by Christopher Armstrong 07 May '02

07 May '02

So, I'm finally on my way to implementing robust PB errors, and I'm getting close to completing phase 1, which is fixing up failure.Failure in preparation for using it for PB errors. Some background: It was decided that we need to be able to trap an error based on a hierarchy; that is, if I try to trap a "WordsError", it should also trap "WordsParticipantError" and "WordsGroupError" (for example). The way to do this is to get a list of all parents of an Exception when a Failure is created with it; I use reflect.allYourBase for this. However, these parents need to be serializable, so the solution is to convert them to strings. The strings obviously need to be globally unique, so I fully qualify them with reflect.qual(). trap(), which is the method one uses for checking wether or not they care about a particular failure, now just checks if any of its arguments are in self.parents. The issue now is what the trap() API should be like. It used to be that fail.trap(types, that, I, want, to, trap) would re-raise the Failure if any of the types matched. This was kinda weird, as it requires code like: try: fail.trap(foo, bar) except: print "Ok, so 'fail' matches 'foo' or 'bar'" Yuck. My proposal is to do this: if fail.trap(foo,bar): #trap() returns self when foo or bar matches print "This is similar to an 'except:' clause" So, if this is acceptable, should I worry about the two cases in Twisted where fail() seems to be called? They don't even seem to make sense in either case (one of them is obviously using trap() incorrectly, the other just seems to use it in an odd way). So can I just change the hardly-used API and fix the current places it's used, or should I remain backwards-compatible? I've attached my diff -u of failure.py, if anyone wants to take a look. It's very small. It makes test_failure fail, because it tests the old behavior of trap(). -- Chris Armstrong << radix(a)twistedmatrix.com >> http://twistedmatrix.com/users/carmstro.twistd/

3 6

[Twisted-Python] zaibach
by j.her＠t-online.de 07 May '02

07 May '02

Hi! If someone edits files in my home dir, please a) ask me b) send me a mail containing WHICH files you edited, and for what reason Thanks. Ciao, Jürgen

1 0

[Twisted-Python] CNAMES and twisted.names in 0.17.3
by drt＠un.bewaff.net 06 May '02

06 May '02

I had some problems with the resolver generating tracebacks for CNAMEs. The decoding has an off by two error which completely messes up the decoding process. The attached patch seems to cure the problem, but I'm not sure why the author of twisted.names did this fiddeling with the offset. Perhaps I'm breaking something else. BTW: The Bugtracker eats my (MacOS) lineendings. drt http://twistedmatrix.com/users/itamarst.twistd/bugs/view_bug?bug_id=12 http://c0re.jp/c0de/misc/Twisted-0.17.3-dns-CNAME.patch diff -ur Twisted-0.17.3-orig/twisted/names/dns.py Twisted-0.17.3/twisted/names/dns.py --- Twisted-0.17.3-orig/twisted/names/dns.py Tue Apr 9 22:37:00 2002 +++ Twisted-0.17.3/twisted/names/dns.py Sat May 4 20:06:51 2002 @@ -156,7 +156,7 @@ if answer.type == 1: answers.append(answer) elif answer.type == dns.CNAME: - answer.strio.seek(answer.strioOff+2) + answer.strio.seek(answer.strioOff) n = dns.Name() n.decode(answer.strio) cnames.append(n.name) @@ -168,7 +168,7 @@ if answer.type == 1: answers.append(cnameMap[name]) else: - answer.strio.seek(answer.strioOff+2) + answer.strio.seek(answer.strioOff) n = dns.Name() n.decode(answer.strio) name = n.name -- teenage mutant ninja hero coders from da c0re - http://c0re.jp/ me - http://koeln.ccc.de/~drt/

2 1

[Twisted-Python] Reactor issues - ongoing
by Itamar Shtull-Trauring 06 May '02

06 May '02

1. Qt and GTK+ assume they deal with processes, in the method unix does it, but this won't work on windows. That is, it is not possible to run processes on windows in Qt or GTK+ event loops at this time. 2. poll() fails on process tests. (3. Qt fails some tests, but only because it doesn't have a working doIteration method, which I will add ASAP.) OTOH, GTK+ and select pass all tests (except an unrelated bug in HTTP code or tests).

1 2

[Twisted-Python] Things to do before next release
by Itamar Shtull-Trauring 06 May '02

06 May '02

1. Move *everything* to new APIs. 2. Add wxdemo.py and qtdemo.py to accepttests. 3. Finish up process refactor so it uses new API. 4. Pass all tests with all event loops. What else? Oh - glyph, start writing ChangeLogs again.

1 0

[Twisted-Python] new reactor testing
by Itamar Shtull-Trauring 05 May '02

05 May '02

I added an option to runtests that lets you choose which reactor to use, e.g.: $ admin/runtests poll or $ admin/runtests gtk The result - the tests only pass in the default reactor, and do *not* all pass in others. So, "Gotta Fix Them All!".

1 0

[Twisted-Python] Kademlia: XOR metric-based routing
by Itamar Shtull-Trauring 05 May '02

05 May '02

Raph Levien says it's less fragile than Chord. Raph Levien is *smart* (http://www.advogato.org/person/raph/diary.html?start=182) <http://kademlia.scs.cs.nyu.edu/> In fact, someone should go read his thesis (/me nominates dash).

1 0

[Twisted-Python] Another gaming position in Austin
by Itamar Shtull-Trauring 05 May '02

05 May '02

<http://www.guru.com/gig/project_view.jhtml?gigId=469455&_DARGS=%2Ffind%2FCM…>

1 0

[Twisted-Python] Interfaces
by Christopher Armstrong 05 May '02

05 May '02

What's up with defining an interface and not specifying any methods on it? from twisted/coil/coil.py: == class ICollection(components.Interface): """A collection for coil.""" == Wow! That's so informative! Now how am I supposed to know which methods I need to define on this class? How do I know what parameters and return values they need? I guess it doesn't matter, I can just go look at the code that uses the objects that conform to the interface, right? PLEASE, if you're going to go to the trouble of defining an interface, *SPECIFY IT*. <fume> -- Chris Armstrong << radix(a)twistedmatrix.com >> http://twistedmatrix.com/users/carmstro.twistd/

1 0