I think it’s tribute to people creating fake Git manual entries with Markov chains and a 17th century forestry book (and the resulting texts being barely distinguishable from the original man pages).
Would it make sense to use concat(sha1, sha256) hash algorithm? This wouldn’t change the prefixes while improving the strength of an algorithm (by including SHA256 in a hash).
No, because trees are hashes-of-hashes, so if you change anything at all about them, their hashes will change, and therefore commit hashes will change.
I don’t think that a different the prefix is a problem anyway. The problem is backwards compatibility and doing a large overhaul of the entire source code, which contains the hashes as hardcoded arrays.
Command line options to convert to new hashes look weird. “Frobnicate blobs”, “climb subtrees”, “use shovels”, “carbon offsets”.
I think it’s intended as a joke/funny example.
Yes. Other parts that made me chuckle:
and
I think it’s tribute to people creating fake Git manual entries with Markov chains and a 17th century forestry book (and the resulting texts being barely distinguishable from the original man pages).
(I can’t find the original “fake” generator, but here is a similar page: https://git-man-page-generator.lokaltog.net/)
https://en.m.wikipedia.org/wiki/Poe%27s_law
I was interested in how fossil handled the SHA1 transition and found it nicely explained here if anyone is interested: https://fossil-scm.org/home/doc/trunk/www/hashpolicy.wiki
Would it make sense to use
concat(sha1, sha256)hash algorithm? This wouldn’t change the prefixes while improving the strength of an algorithm (by including SHA256 in a hash).No, because trees are hashes-of-hashes, so if you change anything at all about them, their hashes will change, and therefore commit hashes will change.
I don’t think that a different the prefix is a problem anyway. The problem is backwards compatibility and doing a large overhaul of the entire source code, which contains the hashes as hardcoded arrays.
I realize there were existing constraints, but this seems like the sort of thing multihash (which I stumbled across from IPFS) would avoid.