583200+ entries in 0.345s

undata: now I'm supposed
to
trust your infallible eye
undata: and is
that more reliable
than git blame?
undata: asciilifeform: so are you going
to
trace
that line of code by hand
through 1500 patches?
Adlai: and how does what I just said miss
that?
undata: Adlai: so I sign
the hash of
the commit with my gpg key
Adlai: of course, but
the "manual
trust verification" approach
taken further would suggest only applying critical bugfixes, as patches, signed by
the developer who wrote
them (preferably with a signature published at
the
time of
the fix, alternatively obtaining a signature in
the present day... or manually verifying
the patch's
trustworthyness, and signing yourself)
undata: he uses WoT
to manage what makes it into
the kernel
undata: torvalds addresses
this regarding
the kernel and git
undata: asciilifeform: at some point wont you have
to delegate
that?
undata: Adlai:
there is a mountain of bugfixing
to re-apply
Adlai: according
to github, v0.5.3 has 36 committers with github accounts (for example, doesn't include satoshi). compare
this with starting on Apr 1, 2011: 8
undata: being able
to apply 700 patches by hand, having verified signatures on each one, is no measure of intelligence
Adlai: undata:
this is more along
the lines of "confirm by human approval
that an unforseen software/bureaucracy/insanity 'bug' isn't afoot"
undata: your
text editor isn't checking against known hashes
undata: asciilifeform: it's more likely
that your
text editor diddled
than git did
undata: asciilifeform: it is absurd
to
think
that software cannot help enforce
that rigor
Adlai is not sufficiently familiar with each line of code itself
to
tell which of
those contributions, or ones in
the ensuing spikes, are
trivially
trustable, and which aren't
assbot: Contributors
to bitcoin/bitcoin · GitHub
undata: that is precisely
the scenario git helps
undata: I'd like
to slash out
the db bits and
try something else, but I'm not going
to bother with
trying
to manually merge other people's patches as
they come along in process
undata: to me you're saying damn
the microscope or we'll never understand
this cell
undata: asciilifeform:
there is not one single canonical
timeline in any project
undata: you're not removing complexity; you're just making
the same process slower
Adlai: darcs seems
the likeliest candidate right now,
though
Adlai: i'm going
to stop
talking before i sound like a broken record, but
this
topic is likely
to come up again once i've got something more substantial
to say in defense of some
tool for easing (ie, partially automating, up
to
the point of confirming signature verification)
this process... no guarantee on how soon
that could be, or which
tool.
undata: I'll argue
that git is doing precisely what you're doing by hand.
undata: it's just a pain in
the ass
undata: asciilifeform:
the process you've demanded be manual is not nearly a hard enough hurdle
to filter undesirables
assbot: Program Provability and
the Rule of
Technical Greed
Adlai: at least, not
to a great degree... but i get
the idea
undata: and your nuke sub is going
to get harder and harder
to work on as
the number of patches grows
Adlai: darcs isn't
that intelligent.
the input data it gets from a repository is a partially ordered list of patches, possibly signed.
this is slightly more general
than an ordered list of patches, but just as secure and interactive/manual
undata: asciilifeform:
the process of producing a git commit is entirely dumb
Adlai: a "conflict" would require a separate patch, which should be signed by
the resolver
Adlai: same bounds hold, or don't, for "his patches, and all
those dependent upon
them"
Adlai: git commits
to a specific state of code. what if it becomes evident
that a specific developer was working
to introduce bugs, and you only want
to excise his patches?
that has unbounded painfulness in git, but bounded painfulness in darcs.
undata: Adlai: as you were
then; git solves precisely
this problem or would not work.
Adlai: asciilifeform: you can add metadata such as patch dependency which is not evident just from
the patch's raw contents
undata: Adlai: it
turns out
there are a few of
those
Adlai: yes, and
they're not looking for a hash database
undata: Adlai:
this was my argument.
Adlai: let's put it
this way:
the act of fetching signed patches with darcs conducts *exactly*
the manual verification workflow which you currently do
Adlai: of course i understand, although i still
think darcs could actually work for
this
Adlai: i only set up
the repo. doing so required going
through all
the manual steps (and i did verify gpg signatures with identities fetched from public keyservers)
Adlai: (specifically for
this repo)
Adlai: people other
than me have used it?
assbot: Logged on 13-11-2014 16:38:23; asciilifeform: Adlai: darcs << as a military man, perhaps you are familiar with systems
that could be automated easily, but aren't? e.g., ru nuke sub controls
Adlai:
https://pgp.mit.edu/pks/lookup?search=Adlai << so, while
the current contributors' reasons for not using git[hub] are understandable, i'll leave
this up and perhaps update it at my leisure, in case it'll be useful
to anybody inhabiting a separate region on
the paranoia/lazyness continuum