< gwillen>
I just learned about Golomb-Rice codes from BIP 158. That's neat.
< gmaxwell>
there are more general/efficient techniques, but golomb coding is at least easy to understand and simple to implement.
< gmaxwell>
(and actually efficient in the unusual case that you can make your probablities just right...)
< sipa>
gmaxwell: you mean range/arithmetic/rANS ?
< gwillen>
the unary quotient/binary remainder scheme is very cute
< gmaxwell>
right (there are another half dozen techniques that also achieve effectively the same thing)
< gwillen>
I am surprised it wasn't presented in the course I had in college where we went over various clever self-delimiting number encoding schemes in an exercise
< gmaxwell>
did you get told the story about how huffman codes were invented?
< gwillen>
we certainly learned about huffman codes, I don't recall a story
< * sipa>
did not learn any stories about huffman
< gmaxwell>
fano basically tricked huffman into solving the problem by presenting it as coarsework, and not mentioning that it was considered a hard open problem.
< sipa>
coarse or course?
< gmaxwell>
of course.
< gwillen>
is there a trick people use when reviewing to see the commits in the proper order? Do you just not use the github interface?
< gwillen>
(github sorts them in a nonsense order, using 'author timestamp' instead of commit timestamp or topological order)
< sipa>
gwillen: yeah, it's a pain and i don't understand why they do that
< sipa>
i complained once and iirc they said they wouldn't fix it
< gmaxwell>
can you use this to make a PR, get some acks on it, then add an out of order commit that appears above the acks?
< sipa>
the overview page always shows things in "edit order"; if you rewrite an earlier commit it gets moved to the bottom etc
< sipa>
so no
< sipa>
it's only in the commit list order (including on the overview page, within groups of multiple simultaneously-pushed commits) that's sorted by author date
< jb55>
then you can git log -p refs/merge/origin/<pr>^..refs/pull/origin/<pr>
< jb55>
I do this call with (magit-log) to do reviews in magit
< jb55>
if you use something like notmuch I have scripts that pulls the PR number automatically out of the subject and launches a magit review: jb55/notmuch-github
< jb55>
only thing that it can't do is code review inside magit, but the magit guy is working on that atm