19:00:26 <wumpus> #startmeeting
19:00:26 <lightningbot> Meeting started Thu Sep 21 19:00:26 2017 UTC.  The chair is wumpus. Information about MeetBot at http://wiki.debian.org/MeetBot.
19:00:26 <lightningbot> Useful Commands: #action #agreed #help #info #idea #link #topic.
19:00:49 <wumpus> #bitcoin-core-dev Meeting: wumpus sipa gmaxwell jonasschnelli morcos luke-jr btcdrak sdaftuar jtimon cfields petertodd kanzure bluematt instagibbs phantomcircuit codeshark michagogo marcofalke paveljanik NicolasDorier jl2012 achow101
19:00:55 <kanzure> hi.
19:00:59 <sipa> hi.
19:01:04 <meshcollider> hello
19:01:09 <cfields> hi
19:01:24 <sdaftuar> hey
19:01:26 <luke-jr> hi
19:01:42 <jnewbery> hello
19:02:06 <achow101> hi
19:02:17 <wumpus> so PSA: we've released with a single patch #11332, compared to 0.15.0, to solve the qt crash issue for some users
19:02:18 <gribble> https://github.com/bitcoin/bitcoin/issues/11332 | Fix possible crash with invalid nCustomFeeRadio in QSettings (achow101, TheBlueMatt) by jonasschnelli · Pull Request #11332 · bitcoin/bitcoin · GitHub
19:02:26 <sdaftuar> yay
19:02:49 <wumpus> #topic high-priority for review
19:03:10 <achow101> #10871 please?
19:03:12 <gribble> https://github.com/bitcoin/bitcoin/issues/10871 | Handle getinfo in bitcoin-cli w/ -getinfo (revival of #8843) by achow101 · Pull Request #10871 · bitcoin/bitcoin · GitHub
19:03:31 <wumpus> I don't think we made much progress on review, this week (me included)
19:03:49 <BlueMatt> no, we (or at least I) did not :(
19:04:23 <sipa> regarding that: i've seen several reports of people wondering what happened because getinfo doesn't work... i guess they went from 0.14 to 0.5.99 and never saw the deprecation
19:04:50 <gmaxwell> Well they shouldn't run 0.5, that would be bad.
19:04:50 <wumpus> added 10871
19:04:51 <BlueMatt> heh, so, what, when a new release is announced they just build master? :(
19:05:07 <luke-jr> Maybe `bitcoin-cli getinfo` should print a message informing the user about the new RPCs, like we did when bitcoind client got deprecated
19:05:12 <gmaxwell> BlueMatt: some people do that.
19:05:15 <wumpus> luke-jr: also thought about that
19:05:19 <BlueMatt> gmaxwell: wat
19:05:26 <luke-jr> lol
19:05:34 <wumpus> luke-jr: it should at least print a deprecated message
19:05:44 <sipa> BlueMatt: gmaxwell is joking about my typo (0.5 instead of 0.15)
19:05:46 <BlueMatt> gmaxwell: so...what you're saying is we should start using a "develop" branch with master pointing to released versions?
19:05:54 <BlueMatt> sipa: i figured......
19:05:54 <wumpus> luke-jr: instead of just about a missing call, as if it's never existed
19:06:21 <sipa> i like the approach being used for the estimatefee deprecation
19:06:23 <luke-jr> BlueMatt: github lets you change the default branch, so we could just make it the latest stable branch
19:06:24 <goatpig> dont poeple typically pull tags, not just the head of a branch (in this case master)
19:06:26 <wumpus> BlueMatt: or we can just change the github default branch to something else, you know
19:06:31 <luke-jr> problem is, it's also the default for PRs…
19:06:34 <BlueMatt> wumpus: yea, possible
19:06:36 <sipa> where it disappears, but you have a cmdline switch to re-enable
19:06:40 <BlueMatt> I mean really not joking there that's a serious issue
19:06:49 <wumpus> but yes, PRs also go there by default, whichi s kinda annoying
19:06:55 <BlueMatt> people building master right after release is usually bad cause we merge lots of fun stuff on master right after release :/
19:06:56 <achow101> topic suggestion: deprecation stuff a la #11031
19:06:57 <gribble> https://github.com/bitcoin/bitcoin/issues/11031 | [rpc] deprecate estimatefee by jnewbery · Pull Request #11031 · bitcoin/bitcoin · GitHub
19:07:05 <luke-jr> I suppose we can expect developers to be smarter than users
19:07:22 <gmaxwell> BlueMatt: I am not saying we should, just pointing out people do that. We have warnings on master... so I don't worry too much.
19:07:22 <cfields> sipa: i like that too. or even an rpc arg like -force-deprecated.
19:07:27 <wumpus> PRs to branches are really annoying, so I'd prefer to just keep the status quo there
19:07:33 <BlueMatt> gmaxwell: yea, maybe we should make the warnings louder?
19:07:37 <BlueMatt> that may be acceptable similar
19:07:38 <sipa> i think the reports  saw were from pretty much first time users "how do i set this up", and following a guide to build from source
19:07:43 <luke-jr> cfields: better to force the user to be involved IMO
19:07:55 <sipa> wumpus: agree
19:08:06 <wumpus> it can't be the release announcement, as I mention the exact tag to check out there...
19:08:13 <jonasschnelli> hi
19:08:15 <cfields> luke-jr: that's what i meant.
19:08:31 <luke-jr> cfields: a RPC arg wouldn't have the user involved
19:08:44 <BlueMatt> maybe bitcoind should need a -iknowthisisunreleasedunstableversion option :(
19:08:46 <luke-jr> software would just set it
19:08:48 <wumpus> maybe would make sense to add a git clone command in there too for people that can't figure that out themselves
19:08:58 <BlueMatt> or configure could take it
19:09:05 <luke-jr> BlueMatt: default to testnet?
19:09:17 <cfields> luke-jr: oh ok, different definitions of user. i see what you mean.
19:09:19 <wumpus> configure already has that information, so that'd make sense
19:09:21 <BlueMatt> ./configure --this-version-is-unreleased-and-possibly-unstable
19:09:37 <BlueMatt> that would make people think twice and make people fix their guides
19:09:43 <luke-jr> s/unstable/insecure/ ☺
19:09:50 <achow101> luke-jr: we would forget to make it default to mainnet for releases
19:10:01 <BlueMatt> heh, ok, at the risk of bikeshedding, I vote cfields implement it...I dunno anything 'bout autotools
19:10:02 <BlueMatt> :p
19:10:04 <luke-jr> achow101: would we? release-process.md
19:10:04 <wumpus> I've never forget to set IS_RELEASE to true for releases
19:10:21 <wumpus> a few times we've forgot to bump the version number but only for *minor* releases
19:10:22 <luke-jr> or just have -testnet default to !IS_RELEASE
19:10:43 <cfields> heh
19:10:46 <wumpus> and for branches the IS_RELEASE is always true anyhow
19:10:49 <BlueMatt> luke-jr: I kinda like it not modifying the bitcoind at all now that i think about it
19:11:01 <sipa> wumpus: an alternative would be to have a bitcoin-releases repo (under the same org), to which only release tags get pushed
19:11:04 <BlueMatt> build the same bitcoind, but make people type something that says "is not release"
19:11:13 <sipa> wumpus: and direct people to clone from there if they just want stable things
19:11:16 <sipa> but meh...
19:11:23 <wumpus> sipa: I don't know...
19:11:29 <BlueMatt> sipa: heh, if we make it not build from bitcoin/bitcoin, ok
19:11:33 <wumpus> I really prefer not to mess with the repopsotiry too much
19:11:44 <jonasschnelli> agree with wumpus
19:11:45 <morcos> This doesn't seem that large a problem to me...
19:11:48 <sipa> wumpus: yes, agree - i don't think we should do anything
19:11:48 <wumpus> agree
19:11:54 <wumpus> any other topics?
19:11:59 <BlueMatt> agreed, though I'd vote for a configure flag
19:11:59 <achow101> we could move the repo to bitcoin-core and break all of those guides :)
19:12:03 <gmaxwell> yea, I think we don't need to do anything except hunt down and kill people with bad guides. :P
19:12:03 <sipa> achow101 had a topic
19:12:06 <BlueMatt> its not a huge issue, but it sounds rather dangerous
19:12:11 <sipa> (though maybe we've covered it already)
19:12:13 <gmaxwell> perhaps increase the profile of tarball downloads.
19:12:18 <BlueMatt> and I dont think hunting down guide-writers is possible
19:12:25 <wumpus> gmaxwell: now that sounds like a good action item
19:12:33 <wumpus> $action hunt down and kill people with bad guides
19:12:44 <gmaxwell> BlueMatt: publishers, not technically authors.
19:12:49 <jnewbery> any other feedback on the approach in #11031? I think it's a useful pattern for deprecating RPCs (which we'll need to do a few times)
19:12:51 <gribble> https://github.com/bitcoin/bitcoin/issues/11031 | [rpc] deprecate estimatefee by jnewbery · Pull Request #11031 · bitcoin/bitcoin · GitHub
19:13:27 <wumpus> #topic deprecation stuff a la #11031
19:13:29 <gribble> https://github.com/bitcoin/bitcoin/issues/11031 | [rpc] deprecate estimatefee by jnewbery · Pull Request #11031 · bitcoin/bitcoin · GitHub
19:13:37 <luke-jr> there's only 363 0.15.99 nodes
19:14:02 <BlueMatt> luke-jr: I'd expect there to be ~0 on mainnet aside from some developer test nodes, fwiw
19:14:18 <achow101> I think that we might run into a problem where people just have -deprecatedrpc (or whatever it is called) and enable all deprecated behavior until it disappears
19:14:43 <BlueMatt> achow101: thats fine, at least they were made to type "I know this is going away soon"
19:14:47 <BlueMatt> so they cant show up and complain that its gone
19:14:48 <cfields> jnewbery: i like it (better than an rpc arg, on second thought), though the name could use some bikeshedding :(
19:15:05 <jonasschnelli> luke-jr: my crawler counts 2496
19:15:16 <jonasschnelli> oh.. no .99,.. sry
19:15:51 <wumpus> I also prefer command line arg to rpc arg
19:15:59 <wumpus> it's enough to type it once
19:16:04 <jnewbery> achow101: it's designed such that the user needs to include `-enablerpcmethod=<method1> -enablerpcmethod=<method2> ...` to avoid the problem of a user just setting `-enablealldepractedrpcmethods` and forgetting
19:16:42 <jnewbery> cfields: yes, name could probably be improved
19:16:45 <gmaxwell> yea, don't have a deprecatedall, it needs to be specific.
19:16:50 <wumpus> the purpose is just to signal, not to overburden a user with extra work updating their client applications (apart from getting rid of using the RPC call in the first place)
19:16:53 <meshcollider> hmm yeah "-enablerpcmethod=" should probably have the word "deprecated" in it
19:17:00 <wumpus> yes
19:17:11 <achow101> I propose calling it -enableddeprecatedrpc
19:17:17 <jonasschnelli> ack
19:17:18 <wumpus> otherwise it sounds more like a silly security feature
19:17:20 <meshcollider> +!
19:17:33 <meshcollider> s/!/1/
19:17:57 <sipa> -yesimnaughtyandneedtoupdatemyclienttonotuserpcmethod=name
19:18:15 <achow101> although maybe enabled may not necessarily be the best word for that since we have PRCs that themselves are not deprecated but have deprecated output fields and only the deprecated output fields would show if that option is set
19:18:33 <luke-jr> thought: should rpcuser auth always throw RPC_METHOD_DEPRECATED? if so, how does the user bypass it?
19:18:43 <morcos> -deprecatedrpc is good enough
19:18:45 <luke-jr> maybe -enabledeprecated= would be better
19:18:51 <wumpus> morcos: yes!
19:18:51 <morcos> in a light pink
19:18:53 <gmaxwell> -backwardscompatibilitylaunchcode=0xdeadbeef234828429342893429  < that could trigger fields that are going away. :P
19:19:05 <wumpus> hehehe
19:19:16 <gmaxwell> (and you need to read the release notes to get the launch code)
19:19:21 <gmaxwell> (or source)
19:19:29 <jnewbery> ok, -deprecatedrpc it is
19:19:35 <wumpus> release notes scavenger hunt
19:19:37 <cfields> gmaxwell: where the launchcode is the git commit, so it changes every day until release
19:20:16 <cfields> jnewbery: ack
19:20:17 <gmaxwell> well the idea there was that each depricated feature would have its own code. So you'd look up the code and specify it.
19:20:19 <luke-jr> gmaxwell: too strong for mere deprecation IMO
19:20:41 <gmaxwell> luke-jr: yes, though it solved the how about deprecated fields case.
19:20:46 <luke-jr> something like that feels more like to allow changing consensus-critical stuff
19:20:55 <achow101> I suppose we can then put all of the deprecated account RPCs behind that argument
19:21:06 <cfields> gmaxwell: you're deprecating me? :(
19:21:11 <wumpus> yes
19:21:14 <luke-jr> gmaxwell: how about if the "launch code" matches the method name for methods? :p
19:21:22 <luke-jr> and "accounts" for accounts
19:21:29 <wumpus> in the case of accounts it'd be a group not one single call
19:21:30 <luke-jr> "rpcuser" for -rpcuser
19:21:42 <wumpus> right
19:21:55 <luke-jr> this code already fits that paradigm I think, except for the param name
19:21:57 <achow101> luke-jr: doing that for rpcuser is going to annoy a lot of people
19:22:16 <luke-jr> achow101: yes, that can be discussed separately; it was an example
19:22:19 <wumpus> I  think we should keep the discussion of what things to deprecate separate
19:22:26 <sipa> agree
19:22:33 <achow101> ok
19:23:56 <jonasschnelli> removing the account support via -depracaterpc and positioned arguments seems pretty difficult and/or dangerous.
19:24:08 <jnewbery> achow101: As long as you're happy with that I'll update 11031 with the new name so you can rebase your various RPC deprecation PRs on it
19:24:18 <luke-jr> "rpc" probably shouldn't be in the arg name.
19:24:43 <jonasschnelli> users may be confuse why account are deprecated but still work while -depacaterpc=accounts
19:24:59 <luke-jr> -enabledeprecated seemed nicer IMO
19:25:05 <achow101> jnewbery: I guess so. I haven't looked over 11031 in a while so I want to review it before rebasing
19:25:12 <wumpus> jonasschnelli: it's just the account rpcs that are deprecated, accounts themselves will work as labels
19:25:16 <meshcollider> jonasschnelli: are you suggesting disable the account system completely?
19:25:23 <meshcollider> (without deprecation)
19:25:48 <jonasschnelli> wumpus: yes. That makes sense.
19:25:52 <wumpus> we'll not rip out the account arguments from any non-account RPC calls
19:26:00 <wumpus> just that move etc are deprecated
19:26:22 <achow101> topic suggestion: what to deprecate
19:26:36 <jonasschnelli> But the transition of having a -enabledepracatedrpc= and not disabling the depracated account features seems not to be ideal.. although I guess it's okay
19:27:02 <wumpus> I doubt there is any ideal way to deprecate a monster like the accounts feature, tbh
19:27:06 <luke-jr> wumpus: getbalance <non-*> should fail too
19:27:35 <wumpus> luke-jr: yes
19:28:13 <luke-jr> jonasschnelli: no "rpc" :/
19:28:17 <wumpus> do we have any more upbeat topic than deprecation?
19:28:37 <gmaxwell> China blocking bitcoin?
19:28:38 <gmaxwell> (not really)
19:28:39 <meshcollider> segwit wallet support progress?
19:28:43 <gmaxwell> That
19:28:52 <sipa> i'll have a PR soon
19:28:52 <wumpus> #topic segwit wallet support progress
19:28:55 <wumpus> thanks
19:29:06 <sipa> otherwise, review #11167
19:29:07 <gmaxwell> sipa: Have you heard from thomasv about them blocking v>0
19:29:09 <gribble> https://github.com/bitcoin/bitcoin/issues/11167 | Full BIP173 (Bech32) support by sipa · Pull Request #11167 · bitcoin/bitcoin · GitHub
19:29:26 <sipa> gmaxwell: yes, we discussed it in here yesterday
19:29:26 <achow101> sipa: how soon is soon
19:29:33 <sipa> achow101: this week, i hope
19:29:58 <jonasschnelli> sipa: are you also tackling the GUI?
19:30:04 <gmaxwell> did you take my advice and call the new file key-i-key-i-o.cpp
19:30:11 <sipa> jonasschnelli: no, i'll need help for that
19:30:21 <sipa> gmaxwell: just key_io
19:30:32 <meshcollider> haha
19:30:35 <jonasschnelli> sipa: please point me to your branch after the meeting so I can start to work on that
19:30:58 <sipa> (for context, gmaxwell is talking about the followup #11372, which is not for 0.15.1)
19:30:59 <gribble> https://github.com/bitcoin/bitcoin/issues/11372 | Address encoding cleanup by sipa · Pull Request #11372 · bitcoin/bitcoin · GitHub
19:31:01 <gmaxwell> There should be almost no GUI intersection, other than perhaps offering the default overrides in the gui.
19:31:23 <sipa> yeah, i expect some expert mode setting to choose the address type - but otherwise it can just use the default
19:31:36 <gmaxwell> oh right, point of use switching, sure.
19:31:49 <jonasschnelli> sipa: yes. That is what I though...
19:32:07 <sipa> not that much to say about the topic otherwise
19:32:14 <gmaxwell> jonasschnelli: another thing in the GUI is that we should eventually have a BIP173 error hinter.  We need a monospace text entry field that can be told to underline characters or something like that.
19:32:25 <gmaxwell> (not necessarily a 0.15.1 feature in any case)
19:32:29 <jonasschnelli> Yes! I just wanted to write that
19:32:53 <gmaxwell> well sipa's bip173 patch doesn't have the error finder in it.
19:33:03 <sipa> i think there is another question, that ThomasV brought up yesterday
19:33:14 <sipa> what to do with private key import/export for segwit addresses
19:33:24 <gmaxwell> Though we've written one (a port of it to JS is on that demo page)... it deserves a nice gui handling though (even the js demo is kind of lame, in that it can't highlight inline)
19:33:33 <wumpus> some importmulti ?
19:33:35 <achow101> sipa: importmulti
19:33:36 <sipa> importmulti can handle this natively
19:33:44 <wumpus> ok, seems good enough for me then
19:33:50 <gmaxwell> importmulti was my first thought too.
19:33:54 <achow101> deprecate the other import* rpcs
19:33:54 <sipa> but perhaps at least a warning for importprivkey / dumpprivkey is needed
19:34:01 <wumpus> just highlight that the old import* rpcs don't work with segwit
19:34:07 <jonasschnelli> sipa: yes. Or depracate?
19:34:11 <gmaxwell> When we do the changes that split the key types later, we'll have more to think about. Right now any key is all keytypes.
19:34:18 <wumpus> or deprecate them at some point, but not for a minor release
19:34:30 <sipa> right
19:34:39 <gmaxwell> we can announce an intent at any time. (I think we just did.)
19:34:46 <ThomasV> importmulti does not seem as easy to use as a serialized format
19:34:54 <wumpus> yes, intent+reason could be menitoned in the release notes
19:35:09 <sipa> ThomasV: i consider that a feature :)
19:35:12 <gmaxwell> ThomasV: because of versioning issues, we're not going to solve a new serialized format right now.
19:35:23 <jonasschnelli> importmulti is a bit cumbersome to use,... but should the okay for an RPC layer
19:35:24 <wumpus> importing keys in bitcoin core is considered advanced functionality anyhow
19:36:02 <sipa> ThomasV: as said, i don't mind discussing a more convenient import format for some use case, but we're not going to solve that instantly
19:36:07 <jonasschnelli> importaddress and key with the current rescan-everything approach must go away at some point anyways
19:36:18 <wumpus> jonasschnelli: yeah... that too
19:36:25 <ThomasV> gmaxwell: we were also considering an import format along the lines of bip124
19:36:28 <gmaxwell> raw key handling frequently results in funds loss, even for advanced users too.. but I do think we should ultimately be embedding the required metadata to actually use a key. but it's not something we can really do right now.
19:36:55 <achow101> the bech32-for-privkeys thing could be something for segwit only and have a the witness version number included in that?
19:37:10 <wumpus> achow101: yes
19:37:25 <sipa> achow101: yes, or no - i'm not sure
19:37:29 <gmaxwell> achow101: witness version number isn't the right data.
19:37:40 <sipa> i think we need a 'pure private key' format, which is just a key
19:37:43 <instagibbs> doesn't tell you how it's being used, specifically
19:37:45 <sipa> to minimize the data needed for backup
19:37:58 <wumpus> more like a 'key use'
19:37:59 <gmaxwell> it effectively needs to communicate the scriptpubkey (perhaps be template reference).
19:38:01 <sipa> the associated addresses/chains/... can be stored separately
19:38:02 <jonasschnelli> Not sure if we should really support exporting / importing private keys over an RPC layer in the long run
19:38:14 <luke-jr> sipa: but if you're going for minimal, you'd backup the seed, not the keys?
19:38:15 <sipa> jonasschnelli: yeah... that's a hard question
19:38:23 <gmaxwell> sipa: still needs metadata.
19:38:29 <sipa> gmaxwell: how so?
19:38:40 <wumpus> right, you could backup the seed, and some metadata, no need to backup the keys themselves
19:38:42 <achow101> <scriptpubkey>|<privkey(s)>
19:39:01 <gmaxwell> I mean you need to know _what_ chains or key types are expected. Otherwise it's a lossy search problem to figure out what you should be actually getting.
19:39:10 <wumpus> yes
19:39:12 <sipa> gmaxwell: sure, but i think that can be separate
19:39:47 <sipa> ideally, your wallet has 1 piece of actually secret data, and then a bunch of chains/addresses/scripts/... that use keys derived from that secret data
19:39:51 <gmaxwell> it's critical data without which you can't recover the keys... and with which you can.  (maybe you could bruteforce it out, under somewhat reasonble assumptions...)
19:40:08 <jonasschnelli> Conceptual, we should work towards private key separation with a daemon (or agent) and core only deals with public keys. The private-keys agent/tool could still be in the repository (or different one)
19:40:12 <gmaxwell> it's also potentially a rather small amount of data.
19:40:30 <sipa> gmaxwell: an example i came up yesterday is CLTV scripts
19:40:43 <sipa> there is no way you can enumerate all CLTV scripts from a particular seed
19:41:02 <gmaxwell> no, indeed, but the inability to back up CLTV cleanly was one of the motivations for CSV.
19:41:16 <sipa> but perhaps there is no need, if you have built-in backups of the chain data (which don't risk monetary loss), and a separate backup of the private key
19:41:30 <luke-jr> so three levels of data really: the seed itself, the types/chains/etc, and the comments/labels
19:41:36 <sipa> the private key in that case is just a piece of secret data, referred to by the chains
19:41:48 <gmaxwell> asking people to handle two critical to maintain pieces of information instead of one would be unfortunate.
19:42:10 <gmaxwell> (and by unfortunate I mean result in non-negligible funds loss in practice)
19:42:24 <luke-jr> gmaxwell: maybe concatenate seed + types/chains/etc in practice
19:42:36 <sipa> sure, for simple situations you can create a convenient format that handles both simultaneously
19:42:38 <gmaxwell> sure. that would be okay to.
19:42:40 <wumpus> you could encode it into one identifier somehow...
19:42:43 <wumpus> sounds easy enough
19:42:45 <gmaxwell> (re to both luke and sipa)
19:42:59 <sipa> but i expect there to be cases where you really just want to backup a key, and will handle the metadata (which is too complicated) yourself
19:43:02 <gmaxwell> the logical seperation sounds fine, but for at least the common case there should be some format for combined data.
19:43:21 <sipa> sure
19:43:22 <morcos> +1 for the common case combined format
19:43:32 <gmaxwell> A possible metadata flag is  'external, good luck'. :)
19:44:13 <jonasschnelli> X509
19:44:20 <sipa> JSONx!!
19:44:21 <ThomasV> lol
19:44:26 * jonasschnelli *ducks*
19:44:40 <wumpus> you could always back up the key and metadata separately, then combine them before importing, or have import accept multiple formats, this is just details
19:44:49 <gmaxwell> Could just have a box where people can just type in x86 machine code and it will decode the result for you ...  almost as flexible as x509.
19:44:57 <gmaxwell> :-)
19:45:10 <luke-jr> XD
19:45:20 <sipa> ok, other topics?
19:45:29 <wumpus> nice, though too easy to use, should be some ancient 8-bit assembly like z80 or 6502
19:46:08 <sipa> INTERCAL
19:46:13 <luke-jr> suggested topic: it would be really nice to get #7533 in, since it is a constant rebase hell otherwise; comments on how to improve the code quality (ugh, macros) appreciated
19:46:15 <gribble> https://github.com/bitcoin/bitcoin/issues/7533 | RPC: sendrawtransaction: Allow the user to ignore/override specific rejections by luke-jr · Pull Request #7533 · bitcoin/bitcoin · GitHub
19:46:16 <gmaxwell> I've had to use 6502 assembly in the last couple years ... (for a mystery hunt puzzle)
19:46:26 <wumpus> yeah
19:47:07 <wumpus> gmaxwell: did you still manage to? :)
19:47:08 <gmaxwell> other topics: did people see on the mailing list that there is a new Dandelion post?
19:47:16 <jnewbery> wumpus: you suggested topic high-priority for review earlier. Did you want to discuss individual PRs or were you just asking if people had PRs they want added to the list?
19:47:38 <meshcollider> https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2017-September/015030.html
19:47:42 <wumpus> #topic Dandelion post
19:48:18 <wumpus> jnewbery: both is ok, though there are too many PRs on the list right now, would prefer some to be reviewed first - the point of it focusing review is lost if we just add all PRs in there :)
19:48:38 <gmaxwell> I don't know that I have much to say, other that I thought it should get some attention (I know not everyone reads the list closely). I think this will be a good thing to implement and the people working on it are pretty eager.
19:48:48 <gmaxwell> (and responsive)
19:49:40 <jnewbery> wumpus: ok, tit-for-tat. I'll review a couple of those in the next week. Can you add #10740. I think it's ready for initial review
19:49:40 <wumpus> thanks for mentioning it, I indeed missed it
19:49:42 <gribble> https://github.com/bitcoin/bitcoin/issues/10740 | [wallet] dynamic loading/unloading of wallets by jnewbery · Pull Request #10740 · bitcoin/bitcoin · GitHub
19:49:55 <wumpus> #link https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2017-September/015030.html
19:50:12 * BlueMatt is curious how this interacts with the questions about mempool sync
19:50:20 <BlueMatt> though I havent read the dandelion stuff much
19:50:43 <wumpus> jnewbery: ok added
19:50:43 <gmaxwell> BlueMatt: externally to it. Basically it's a quiet forwarding thing that has little to no mempool interaction.
19:50:53 <jnewbery> wumpus: thanks!
19:50:58 <gmaxwell> BlueMatt: so that public distribution of transactions happens away from the source.
19:51:36 <BlueMatt> gmaxwell: well based on the dandelion tl; dr i just got, it seems to me that dandelion is essentially the old mempool-sync proposal of "relay txn to only one, maybe two peers but then do sync of your mempool with your peers every so often to propagate things more fully"
19:51:48 <BlueMatt> except they replace sync with "broadcast after a while"
19:52:07 <gmaxwell> BlueMatt: kind of. it's not in your mempool though when it passes through in the one at a time propagation.
19:52:35 <sipa> BlueMatt: but dandelion has rebroadcast too, which can't be avoided (if you dandelion-forwarded a TX, but don't see it N seconds later on the normal network, you yourself start broadcasting it normally)
19:52:57 <morcos> sipa: why not dandelion it again?
19:53:23 <BlueMatt> sipa: yes, I'm asking for comparison between dandelion and the old mempool-sync proposal of gmaxwell
19:53:34 <sipa> morcos: the dandelion stem isn't expected to reach the entire network
19:53:54 <BlueMatt> it sounds to me like the gmaxwell proposal is effeciely similar, though maybe dandelion pointed out (correctly, I suppose) that you want to not add it to your mempool until later to preserve privacy
19:53:56 <sipa> so you still need something outside of dandelion-forward and mempool reconciliation
19:54:00 <gmaxwell> Peer hands you a txn with a private flag. Assuming the txn is valid wrt your mempool, you set it aside and forward it on to a single other peer (determinstically selected based on the input peer).  At random, you drop the private flag when you relay it with some fixed probablity.   If the txn is offered by other peers you fetch it.
19:54:36 <gmaxwell> If after a timeout you never see it from other peers you broadcast it yourself.. but that only happens if someone in the 'stem' path drops the ball.
19:54:56 <sipa> BlueMatt: there seems to be some overlap, but i don't think dandelion+sync is a complete solution
19:55:07 <gmaxwell> so it is similar to my relay improvement proposal, but the privacy requirements mean that it doesn't accomplish syncing itself.
19:55:12 <morcos> sipa: hmm, i was just asking if you didn't see it, why you wouldn't repeat the whole process assuming the stem got cut somehwere, and hoping that you can get to the fluff on try 2.  why fall back to old method
19:55:39 <BlueMatt> wait, wait doesnt "(determinstically selected based on the input peer)" imply that you can still group transactions from a single source based on where you first saw them....sure where you first saw them wont be the guy who created it, but it'll still be the same for all txn from the same guy
19:55:44 <BlueMatt> maybe I should shut up and go read it
19:55:44 <sipa> morcos: the 'fluff' is just normal tx relay
19:55:48 <gmaxwell> BlueMatt: so it is like my relay bandwidth improving proposal, without improving relay bandwidth. :P
19:56:20 <BlueMatt> heh, yea, ok
19:56:28 <gmaxwell> BlueMatt: only if you are inside the stem yourself.  (this was a change their latest work makes)
19:56:38 <BlueMatt> hmm, alright
19:57:24 <gmaxwell> the tradeoff is that if you don't do the determinstic routing and the sender sends lots of txn you are certian to be in the stem for some of them.
19:57:49 <gmaxwell> basically related to the guardnodes problem in tor.
19:58:38 <gmaxwell> in any case, it would be good to have more critical eyes on their design. I've only just started thinking about the implications of the change to determinstic paths.
20:00:01 <sipa> PLOINK
20:00:11 <wumpus> yes, extrememly important for privacy to not make any mistakes here
20:00:18 <wumpus> #endmeeting