Time  Nick        Message
00:07 hmmmm       as for the mapgen flags for creating the world
00:07 hmmmm       light should not be an option there
00:07 hmmmm       nobody is ever going to want to explicitly set that
00:08 hmmmm       it's for modmaking use exclusively
00:09 hmmmm       there's also the deprecated v5 flags 'blobs'
00:10 hmmmm       i feel as if the GUI code doing this is very repetitive
00:10 hmmmm       this guy could use loops
01:14 Yepoleb     hmmmm: i found an irc log where you were talking about client-side lua, what's the current state of it?
01:14 hmmmm       design and planning
01:15 Yepoleb     it ended with a discussion about running non-free code and signing
01:16 Yepoleb     did you find a solution for that?
01:19 hmmmm       i'm not absolutely sure if i had discussed this previously, but we can't force client-side mods to have any particular license
01:19 hmmmm       signing is doable although impractical - this is akin to running javascript on a particular webpage
01:28 Yepoleb     hmmmm: i think if it is completely sandboxed, you don't need any additional security
01:29 hmmmm       sure.. there are exploits though
01:30 Yepoleb     what kind of exploits?
01:30 hmmmm       breaking out of the sandbox, running arbitrary code on the host machine
01:32 Yepoleb     yeah, but what would you explot?
01:34 hmmmm       you don't know until it's too late
01:38 Yepoleb     but you also can't secure against an unknown attack
01:55 Yepoleb     hmmmm: browsers just have a sandbox and it seems to work quite well
01:55 hmmmm       are you kidding me
01:55 hmmmm       there were so many heap spraying exploits
01:57 Yepoleb     how would you prevent exploits?
01:57 hmmmm       we can't, but that's why code signing is so appealing
01:58 hmmmm       the lua execution environment is a very large surface area for exploitation
02:06 Yepoleb     hmmmm: yeah, but signing would be way too much work
02:07 Yepoleb     and you can't just implement a quick bugfix, because it would take a few days to get signed
02:10 Yepoleb     server owners wouldn't be able to modify anything
02:15 ShadowNinja Um, please *do NOT bundle GMP*.  Running `pacman -S gmp` is cool, downloading an extra hundred megs which will always be in the Git history bloating the repo and extending the compilation step by a few minutes building GMP is *definitely NOT cool*.  And all packagers are going to specifically avoid anything packaged for a number of good reasons.
02:15 ShadowNinja s/packaged/bundled/
02:15 hmmmm       keep reading
02:17 paramat     i added comments to #2561 as i'm unsure if this PR works correctly with minetest.conf and minetest.conf of the subgame
02:17 ShadowBot   https://github.com/minetest/minetest/issues/2561 -- Add mapgen specific flags to create world dialog by srifqi
02:18 hmmmm       i'm trying to come up with some way to deprecate the flag system
02:18 hmmmm       it wasn't a very smart idea to expose that implementation detail to the end user
02:19 hmmmm       that really should be a set of boolean settings
02:20 ShadowNinja mini-gmp sounds slightly better, but we shouldn't bundle anything IMO, you should always be using the full GMP anyway.
02:21 paramat     i have some corrections to conf.example to do, while i'm at it i will move the jungles and dungeons flags from mtgame into the engine, hope this is okay..
02:21 ShadowNinja The only reason JSONCPP's still in the tree is because some distros' JSONCPP cause segfaults when linked against Minetest.
02:22 ShadowNinja And the only other bindled lib is Lua.  And that's causing some major issues for the IRC mod...
02:22 ShadowNinja (bundled Lua isn't configured with the distro's search paths)
02:24 hmmmm       what is the argument against having default bundled libraries
02:25 ShadowNinja hmmmm: They're not configured for the distro, they can't be updated without rebuilding all packages that bundle them, they waste memory because each version has to be loaded...
02:26 ShadowNinja I'm sure you can find a dozen other reasons with some searching.
02:26 hmmmm       why would they need to be configured for the distro?
02:26 hmmmm       minetest is only bundling a couple libraries, and this is something that does not need to be updated often
02:27 hmmmm       in fact we're using an older version on purpose for freebsd compatibility
02:27 hmmmm       third, they don't waste any memory because we'd simply *not* link against that object file if it has been chosen to use the system installation
02:28 hmmmm       you do understand that mini-gmp is a single C source file, correct?
02:28 hmmmm       it's highly portable and by no means forced onto the package builder.  they can easily use the system version instead, in fact which is recommended for servers
02:29 hmmmm       on the other hand, bundling a default library saves a lot of pain and trouble that you'd otherwise experience with dependencies
02:32 ShadowNinja hmmmm: The bindled Lua only has /usr/local in its package.(c)path, but most distros use /usr.  The IRC mod comphensates for this with a hack, but it still fails on a lot of distros that use other search paths (eg, in /opt or /usr/lib-x86_64-Linux-GNU).
02:34 ShadowNinja We can be compatible with (old) FreeBSD by simply not using 6.0-only features and letting users install whatever is the latest version available to them.
02:35 hmmmm       not old freebsd, versions as recent as 9.2
02:35 ShadowNinja mini-GMP might be OK, but I'd prefer a simple dependency.
02:35 hmmmm       simply put, libgmp tests the limits of many compilers and consequently breaks them
02:36 ShadowNinja hmmmm: 9.3 and 10.1 are out, so yes, it's old.  Maybe not obsolete, but old.
02:36 hmmmm       GNU libraries also arrogantly make windows a second-class citizen.  i am concerned with depending on GNU libraries especially for this reason
02:36 hmmmm       when making my own windows build, I've noticed I had the most trouble with the GNU libraries
02:37 ShadowNinja That might be a problem.
02:37 hmmmm       the bundled lua having only /usr/local as its package.cpath is a problem, but that's a bug with lua.  i wouldn't say that's a reason to not pacakge it
02:37 ShadowNinja I consider Windows a "second-class citizen" too, but I still recognize that it's important due to its userbase.
02:37 hmmmm       right
02:38 hmmmm       that same mentality carries over onto clang though
02:38 hmmmm       when I see that "freebsd 10.x x86 is broken" and so on it raises my concern
02:38 hmmmm       they don't care to work around known problems
02:38 ShadowNinja hmmmm: No, Lua can't include every path that every distro uses.  Heck, a user might decide to use somwhere in /home.
02:39 hmmmm       in general, dependencies are made easy to install with linux/bsd/etc.
02:40 hmmmm       dependencies are serious business and we can't expect everybody to have 100% access to them, which is why I'd rather have a builtin version
03:49 paramat     i would like to push this later #2623 back in ~1hr
03:49 ShadowBot   https://github.com/minetest/minetest/issues/2623 -- Conf.example: Update, fix and improve mapgen flags docs by paramat
05:03 hmmmm       hrmm
05:04 hmmmm       darnit where'd paramat go
05:19 paramat     here
05:20 paramat     i'll wait for you to push 'serialize schematic' first, to avoid conflicts
05:30 hmmmm       well
05:30 hmmmm       i don't think you'd like it right now
05:31 hmmmm       if you were to call serialize_schematic("foobar.mts", "lua") you'd get a table with untranslated node IDs
05:31 hmmmm       (well, if you were to call that during mod initialization, i mean to say)
05:32 hmmmm       the solution I am coming up with in another commit, to introduce the concept of the "node resolution strategy"
05:32 hmmmm       which you'll be able to globally set for all node resolving operations, or specifically during serialize_schematic
05:32 paramat     actually there are no conflicts with my PR 2623 =)
05:33 hmmmm       three node resolution strategies:  NODE_RESOLVE_DIRECT, which directly looks up the content ID based on the contents of the name table currently existing, which means that node resolution may fail if a certain mod it contains nodes from hasn't been loaded yet
05:34 hmmmm       then NODE_RESOLVE_DEFERRED, which pends node resolutions until the registration phase has ended at which point it goes back and resolves all the nodes used in the schematic
05:34 hmmmm       and then a third, NODE_RESOLVE_NONE, where it stores a direct string list of what was loaded from the schematic definition (table or mts file, doesn't matter) and doesn't attempt to perform any resolution whatsoever
05:35 hmmmm       for schematic serialization, you'd want to use NODE_RESOLVE_NONE no doubt and then perhaps there should be a flag to not auto-register the schematic upon loading
05:41 paramat     i see. personally all i need a the moment is to (within a mod for the purpose) define a schem using a lua table, and have a way to save that as a .mts file
05:53 paramat     i'll push #2623 in a few minutes
05:53 ShadowBot   https://github.com/minetest/minetest/issues/2623 -- Conf.example: Update, fix and improve mapgen flags docs by paramat
05:56 hmmmm       paramat
05:57 hmmmm       that's not just a doc update, you slipped in new default parameters
05:57 hmmmm       that would break worlds
06:02 paramat     ah..
06:02 hmmmm       here's the way flags work:
06:03 hmmmm       there is always a baseline, default parameter hard coded into the engine
06:03 hmmmm       flags loaded out of config files or modified in lua are differences to that default parameter
06:03 hmmmm       so that way you don't need to specify "light" in mg_flags in the config
06:03 hmmmm       meaning that if people didn't explicitly put jungle, or dungeons, it would start doing so
06:04 paramat     oh man i didn't see that
06:04 hmmmm       same with worlds that were already created
06:04 paramat     will remove that stuff then, doc update only
06:04 hmmmm       map_meta.txt contains all flags available written in their explicit forms, but if a new default were added then it'd activate it by default in those map_meta.txts
06:05 hmmmm       so then you'd get inconsistent worlds
06:07 paramat     thanks for the review, updating PR..
06:22 paramat     updated, will push when checks are done
06:33 paramat     now pushing 2623
06:38 paramat     done
10:26 est31       about mini-gmp vs. hard dependency: I think soft dependencies are better than hard ones. git can manage that additional 4k lines file.
10:31 est31       we should urge distro packagers to depend on real libgmp though. mini-gmp is clearly inferior.
12:37 cd2         hi
17:02 ShadowNinja Any objections to removing the "simple menu" that's used for Android?  It prevents you from having more than one world and has a lot of code duplication.
17:03 rubenwardy  Would that remove touch friendliness?
17:13 ShadowNinja rubenwardy: No, it would just give a slightly different menu layout.
17:14 rubenwardy  Okay
17:15 rubenwardy  My phone is terrible, doesn't have enough memory. I recommend you stay away from Huawei, their phones and customer service are/is crap. </partially-offtopic-rant>
17:19 Calinou     Backdoorhei
17:19 Calinou     (this is widely known)
17:19 rubenwardy  Lol
17:22 Hijiri      what does minetest use gmp for?
17:25 Calinou     authentication, IIRC
17:26 Hijiri      oh
17:47 ShadowNinja It doesn't use GMP yet, it may be added in the future though.
17:50 Hijiri      oh x 2
17:50 Hijiri      thanks guys
18:39 nrzkt       ShadowNinja: agree for the unification of Android and PC platforms, Android menu is not pertinent.
18:56 * est31     tests whether hmmmm is right and minetest doesnt compile for c++11...
18:58 rubenwardy  We should take ideas from HTML's responsive layouts
18:58 rubenwardy  Although CSS is horrible and fiddly
19:03 est31       hmmmm is right
19:03 est31       aaand its inside jsoncpp...
19:04 hmmmm       hrm
19:07 est31       .git/rebase-apply/patch:5429: new blank line at EOF.
19:07 est31       very easy to track down lol
19:07 est31       (trying SN's patch)
19:10 est31       but it seems only to be a problem with clang
19:10 est31       http://reviews.llvm.org/rL201729
20:43 paramat     hmmmm i just realised *=/ i can move the jungles and dungeons flags from mtgame's .conf into the engine's .conf, i guess this way is okay?
21:18 hmmmm       yse, that would work fine
21:20 paramat     good, will prepare a PR
22:01 paramat     okay will push #2625 later if approved
22:01 ShadowBot   https://github.com/minetest/minetest/issues/2625 -- Defaultsettings: Add dungeons and jungles flags. Update conf.example by paramat