[tahoe-dev] Weekly Dev Chat notes, 2012-12-13

[Zooko Wilcox-O'Hearn]

Folks:

We decided not to do the hangout "on air" this time, as some people
feel like they would be inhibited in their conversation if it were
recorded and published like that. Maybe we could try an experiment in
putting it "on air" at some point.

Regards,

Zooko

in attendance: Zooko (scribe), Marlowe, CodesInChaos, David-Sarah, Brian

On internationalization: marlowe is following up with runa to get
instruction in how the workflow works for the Tor project and set up
the same thing for the Tahoe-LAFS project; a man whose name I didn't
catch will hopefully volunteer to translate it into Arabic. Brian will
ask around Mozilla how they do it and Brian and Marlowe can compare
notes. (Tor and Mozilla are two projects that do this sort of thing
quite successfully.)

* Zooko wants to get #1679 closed. It is a critical issue for those
that it strikes (which includes LeastAuthority.com customers), and
there is a patch. We just need a unit test. Zooko started writing unit
tests for NodeMaker during the call. Brian found some extants tests
for NodeMaker, in test_client.py.

* marlowe is working on documentation improvements, hopefully to go
into the 1.10 release.

* marlowe is starting a glossary. He'll maintain it on the wiki, but
then we'll copy a snapshot of it into the source release. Marlowe
figured out that you can put restructured-text into trac wiki with
"{{{#!rst". Ideally, we maintain restructured-text formatted glossary
in the wiki, and then copy exactly the same restructured-text file
into the docs/ directory before making a source release.

Wiki vs. revctrl ? We've traditionally wanted to put docs in revision
control to replicate them to users, make them linked to the version of
the source code that they come with, and maintain a useful history of
the docs. But we've also traditionally wanted to put docs on the wiki
in order to make it easier for people to edit them. Maybe github will
satisfy both goals! Brian pointed out that github is introducing an
"edit right here on this page" feature so that it is even easier for
people to contribute patches.


Brian showed up late, but we wanted to talk to him, so we had an extra
long call.

* CoolNewHashFunction to be announced soon! The pitch is that it is a
modern, secure hash, comparably secure to SHA-3, but it is faster than
MD5. On the best-suited platform, with parallelized computation and
the wind at your back, it might be up to 10 times as fast as SHA-256!
Brian pointed out that this might not make any difference -- secure
hashing is probably not the bottleneck in Tahoe-LAFS. Zooko laughingly
countered that this is because our network protocol is so inefficient.
Brian said we have some measurements of encryption and erasure coding
(see "Recent Uploads and Downloads" in the WUI) but not of hashing
because hashing gets interleaved with other operations so it isn't
that easy to measure. Zooko said there was a cool Master's Thesis by
Eirik Haver and Pål Ruud in which they attempted to measure the effect
of the secure hash function on Tahoe-LAFS's performance:

https://tahoe-lafs.org/trac/tahoe-lafs/wiki/Bibliography#HashFunctions


Brian volunteers to be Release Manager for Tahoe-LAFS v1.10.0! Hooray!
There was much rejoicing!

* There are a few other tickets that we *really* want to get in --
#1240, #1732, and #1767 -- either because they are already done and
just need to be merged or because they threaten forward-compatibility
issues if we don't fix them before the release.

Next week's Tahoe-LAFS Weekly Dev Chat will be about Tahoe-LAFS v1.10.
Be there or be square!