| Time |
Nick |
Message |
| 00:50 |
MTDiscord |
<Jonathon> not that i know of |
| 00:50 |
MTDiscord |
<Jonathon> you could make a mod that teleports them on login |
| 00:54 |
Alias2 |
Or you could edit the database where their logout position in saved |
| 01:01 |
|
Tarap joined #minetest |
| 01:06 |
|
calcul0n joined #minetest |
| 01:21 |
|
absurb joined #minetest |
| 01:29 |
|
Verticen joined #minetest |
| 01:56 |
|
illwieckz joined #minetest |
| 02:21 |
|
Hawk777 joined #minetest |
| 02:26 |
|
YuGiOhJCJ joined #minetest |
| 02:28 |
|
queria joined #minetest |
| 02:33 |
|
queria joined #minetest |
| 02:40 |
|
illwieckz joined #minetest |
| 03:28 |
|
kamdard_ joined #minetest |
| 03:42 |
|
dzho joined #minetest |
| 04:00 |
|
MTDiscord joined #minetest |
| 04:35 |
|
Kimapr[i] joined #minetest |
| 04:38 |
|
dzho joined #minetest |
| 05:01 |
|
riff-IRC joined #minetest |
| 05:10 |
|
MTDiscord1 joined #minetest |
| 05:12 |
|
fkbm joined #minetest |
| 05:21 |
|
sagax joined #minetest |
| 05:30 |
|
TomTom joined #minetest |
| 06:05 |
|
CWz joined #minetest |
| 07:04 |
|
ssieb joined #minetest |
| 07:14 |
|
specing_ joined #minetest |
| 07:47 |
|
Markow joined #minetest |
| 08:03 |
|
hendursa1 joined #minetest |
| 08:03 |
erlehmann |
sfan5, FYI: https://git.minetest.land/MineClone2/MineClone2/issues/1846 |
| 08:04 |
erlehmann |
sfan5, regarding “point them at the packet counter feature”, less lag and less crashes are the main reason mineclonia exists |
| 08:05 |
erlehmann |
everything else (better banners) is just something nice to have |
| 08:06 |
|
hendursa2 joined #minetest |
| 08:07 |
mazes_83 |
about detached inventory id, I see people maintain manufacturing_number, would it be reasonable to use: minetest.get_current_modname() .. ":" .. string.format("%05x-%05x-%05x-%05", math.random(1048575), math.random(1048575), math.random(1048575), math.random(1048575)) |
| 08:09 |
mazes_83 |
that would give a 16^20 collision space, and could save a table for current mod to check if the string is already in use |
| 08:10 |
erlehmann |
mazes_83 what exactly are you trying to do here |
| 08:11 |
mazes_83 |
generate a unique id for detached inventories |
| 08:12 |
erlehmann |
mazes_83 be more specific please. why would you want to do that? |
| 08:13 |
mazes_83 |
from what I understood detached name have to get a unique id ? |
| 08:13 |
mazes_83 |
s/name/inventory |
| 08:14 |
mazes_83 |
I see in the mod I edit, maintains a table on disk to get the next free id |
| 08:15 |
mazes_83 |
I just try to look for another way |
| 08:17 |
erlehmann |
why would you want to do that? |
| 08:18 |
erlehmann |
why detached inventories |
| 08:18 |
erlehmann |
for what |
| 08:19 |
mazes_83 |
I see people adding features on the mod I edit introduced errors on this disk table, so I try to find another way |
| 08:19 |
mazes_83 |
detached inventories are per entities this case |
| 08:19 |
mazes_83 |
per entity |
| 08:31 |
sfan5 |
erlehmann: good to know that MCL2 devs are aware and I don't need to feel bad for not bothering to open an issue |
| 08:32 |
sfan5 |
however the "just move it to the client" stance seems short-sighted. even if you move the code there the data still has to get there, at which point you have to solve an identical problem |
| 08:41 |
erlehmann |
sfan5 knowing and caringn are different things sadly. they sometimes port the fixes from mineclonia, after months. |
| 08:43 |
erlehmann |
sfan5 the “just move it to the client” thing bothers me more because someone who writes code that updates a texture 66,6 times per second on the server will do the same thing on the client. the result: lag. |
| 09:03 |
|
olliy joined #minetest |
| 10:08 |
|
Kimapr[i] joined #minetest |
| 10:26 |
|
calcul0n joined #minetest |
| 10:37 |
|
tech_exorcist joined #minetest |
| 10:57 |
|
riff-IRC joined #minetest |
| 10:58 |
|
IvanGorinich joined #minetest |
| 12:14 |
|
calcul0n_ joined #minetest |
| 12:33 |
|
Kimapr[i] joined #minetest |
| 12:44 |
|
CWz joined #minetest |
| 12:47 |
|
fkbm joined #minetest |
| 12:48 |
mazes_83 |
erlehmann: what are main differences between "mineclone 5" and mineclonia ? |
| 12:52 |
erlehmann |
mazes_83 mineclone 5 was made because new features were not added. mineclonia was made because of the horrible lag that mineclone 2 (and mineclone 5) have and bc the devs kept adding new bugs. |
| 12:52 |
erlehmann |
for example, in mineclone 2 and mineclone 5 for months now all mobs that are not in extra_mobs have regressions compared to mineclone2 0.71. |
| 12:53 |
erlehmann |
like, missing AI and stuff |
| 12:54 |
erlehmann |
mazes_83 naturally, mineclonia gets new features slower than mineclone 5. but you can, for example, run mineclonia and install the copper mod i think. |
| 12:54 |
|
mary4 joined #minetest |
| 12:57 |
erlehmann |
mazes_83 if you play singleplayer, the lag will most likely not be noticeable. for multiplayer, i'd suggest to figure out if you want to run mineclonia. it will not have the new nether mapgen ofc. |
| 12:58 |
erlehmann |
the broken mobs are a bit frustrating. they can wear armor though, that's a plus lol |
| 13:09 |
|
kamdard_ joined #minetest |
| 13:13 |
|
hendursaga joined #minetest |
| 13:30 |
|
Wuzzy joined #minetest |
| 13:31 |
IceDragon |
mazes_83: For unique ids, I would suggest something like ulid https://github.com/ulid/spec |
| 13:31 |
IceDragon |
Granted you would need a stable time source of some kind |
| 13:32 |
IceDragon |
Maybe a globalstep service that just increments a counter and is saved to disk somewhere so it survives world reloads |
| 13:32 |
IceDragon |
Since the system time can drift |
| 13:37 |
|
DrFrankenstone joined #minetest |
| 13:39 |
sfan5 |
just appending a random number to the current gametime seems decent |
| 13:44 |
IceDragon |
As long as its monotonic any time would work methinks |
| 13:45 |
|
Taoki joined #minetest |
| 13:51 |
|
peterz joined #minetest |
| 13:54 |
mazes_83 |
sfan5: how to get the gametime, this is the kind of entropy I need (despite the 16^20 space I use) ? |
| 13:57 |
mazes_83 |
for now I just use strformat("%s:%x-%x-%x-x", minetest.get_current_modname(), math.random(1048575), math.random(1048575), math.random(1048575), math.random(1048575)) |
| 13:58 |
|
Verticen joined #minetest |
| 13:58 |
mazes_83 |
I also maintain a table of already used "seed", to guarantee my seed hasn't already been used |
| 13:59 |
IceDragon |
`minetest.get_gametime()` |
| 14:01 |
mazes_83 |
oups, I could have find this alone, sorry |
| 14:02 |
IceDragon |
With ULIDs unless you're somehow generating billions of ids per second, you should never run into conflict |
| 14:02 |
IceDragon |
So it can eliminate the 'seen' or seed table |
| 14:03 |
IceDragon |
If it brings you any comfort, it's the id format I use at work for messaging and cdrs |
| 14:06 |
mazes_83 |
I think I'll stick to the simple solution I found, I think it's also kind of robust to collision |
| 14:06 |
|
calcul0n joined #minetest |
| 14:53 |
|
Fixer joined #minetest |
| 15:09 |
|
mary4 joined #minetest |
| 15:15 |
Fixer |
sfan5: thaaaaaaaaaanks |
| 15:40 |
erlehmann |
sfan5, mazes_83 how random is math.random btw? |
| 15:41 |
Krock |
minetest initializes the seed somewhere in builtin, so it's okayish random |
| 15:44 |
erlehmann |
can i steal the inv somehow? |
| 15:44 |
erlehmann |
if i know the id |
| 15:53 |
|
absurb joined #minetest |
| 16:16 |
|
Extex joined #minetest |
| 16:25 |
mazes_83 |
yes ofc, but id were even more easy to guess before this random |
| 16:27 |
mazes_83 |
they just followed this pattern: entity_name .. ":" .. manufacturing_number |
| 16:27 |
mazes_83 |
were manufacturing number is a sequential |
| 16:30 |
mazes_83 |
I should put the id table to local instead, would be more sane |
| 16:35 |
|
delta23 joined #minetest |
| 17:03 |
|
calcul0n_ joined #minetest |
| 17:08 |
MTDiscord |
<luatic> Krock: Yes, Minetest seeds the random. But it uses os.time(), meaning servers started the same second will have the same sequence. Minetest does not pop a few random numbers, leading to rather poor first random numbers. |
| 17:13 |
|
debiankaios joined #minetest |
| 17:18 |
IceDragon |
Any chance of adding a higher precision timer to the lua api so we can at least get millisecond accurate time? |
| 17:19 |
IceDragon |
That would also allow for a better random seed |
| 17:19 |
|
Mineclone2coolga joined #minetest |
| 17:20 |
Mineclone2coolga |
I also developed Ark survival evolved for minetest very big game because 3 giga but I don't know how to post it how to do? |
| 17:20 |
MTDiscord |
<luatic> Mineclone2coolga: Troll somebody else. |
| 17:20 |
MTDiscord |
<luatic> IceDragon: I use minetest.get_us_time() which, as the name suggest, offers microsecond precision |
| 17:21 |
MTDiscord |
<luatic> For the seed, you can alternatively add os.time() and os.clock() |
| 17:21 |
|
Talkless joined #minetest |
| 17:21 |
sfan5 |
IceDragon: good suggestion, in fact so good that such a function was added more than 3 years ago |
| 17:21 |
sfan5 |
;) |
| 17:21 |
Mineclone2coolga |
[19:10:59] <Mineclone2coolga> this ark minetest it contains all the maps such as |
| 17:21 |
Mineclone2coolga |
[19:10:59] <Mineclone2coolga> the island |
| 17:21 |
Mineclone2coolga |
[19:10:59] <Mineclone2coolga> scorched earth |
| 17:21 |
Mineclone2coolga |
[19:10:59] <Mineclone2coolga> aberration |
| 17:21 |
Mineclone2coolga |
[19:10:59] <Mineclone2coolga> extinction |
| 17:21 |
Mineclone2coolga |
[19:10:59] <Mineclone2coolga> genesis part 1 |
| 17:21 |
|
Mineclone2coolga was kicked by ShadowBot: Message flood detected. Use a pastebin like paste.ubuntu.com. |
| 17:21 |
|
Mineclone2coolga joined #minetest |
| 17:21 |
IceDragon |
How did I miss that looking over the lua api, I feel stupid |
| 17:21 |
MTDiscord |
<luatic> If you really did build such a large game, you would sure know how to publish it. |
| 17:22 |
Mineclone2coolga |
no |
| 17:22 |
MTDiscord |
<luatic> But in case you have done something, perhaps built maps: Either find a way to reduce your file size, or find a free hoster which accepts such large files. |
| 17:29 |
|
CWz joined #minetest |
| 17:32 |
Mineclone2coolga |
because minetest will not hold all this? |
| 17:32 |
MTDiscord |
<Sublayer plank> well minetest doesn't have built-in file hosting capabilities |
| 17:33 |
MTDiscord |
<Sublayer plank> unless... you'd host a server from it |
| 17:33 |
MTDiscord |
<Sublayer plank> but if you want others to download it and play it locally you can't do it by hosting a minetest server |
| 17:34 |
Mineclone2coolga |
ok bye |
| 17:34 |
|
Mineclone2coolga left #minetest |
| 17:41 |
erlehmann |
IceDragon you won't get secure random numbers out of minetest ig |
| 17:45 |
|
erle joined #minetest |
| 17:46 |
|
appguru joined #minetest |
| 17:55 |
|
Markow joined #minetest |
| 17:58 |
|
mary4 joined #minetest |
| 18:05 |
|
appguru joined #minetest |
| 18:08 |
|
TomTom joined #minetest |
| 18:08 |
IceDragon |
erlehmann: My comment on a better random seed was in response to Krock mentioning the use of os.time which is only accurate up to a second |
| 18:09 |
IceDragon |
If I really wanted strong random numbers I would load up openssl through ffi |
| 18:10 |
erlehmann |
any use of os.time() as the sole seed is ridiculous |
| 18:10 |
sfan5 |
no need for openssl either https://github.com/minetest/minetest/blob/6fedee16f098549ffaee188b02b777239513abc3/doc/lua_api.txt#L6986-L6997 |
| 18:13 |
IceDragon |
I really need to look through the lua_api more |
| 18:13 |
IceDragon |
I completely forgot about SecureRandom |
| 18:14 |
|
sec^nd joined #minetest |
| 18:15 |
MTDiscord |
<luatic> Wait we have secure random? |
| 18:15 |
MTDiscord |
<luatic> The Minetest feature tree is a very odd one |
| 18:15 |
sfan5 |
since 2015-08-06 yes |
| 18:17 |
IceDragon |
the minetest api has all kinds of nice things, they're just tucked away in corners of the document sometimes |
| 18:17 |
|
sec^nd joined #minetest |
| 18:39 |
|
sec^nd joined #minetest |
| 18:46 |
|
appguru joined #minetest |
| 18:55 |
|
Flabb joined #minetest |
| 18:57 |
|
longerstaff13 joined #minetest |
| 19:00 |
IceDragon |
So I was messing around with ItemStack, I did a benchmark, where I created 4 million item stacks, it took 30+ seconds, if I ditch the constructor and leave just a basic table ({name, count, wear}) it takes 1 second for all the stacks |
| 19:01 |
IceDragon |
Now what I'm wondering, what exactly goes on in the ItemStack constructor that would make it that slow |
| 19:13 |
|
specing_ joined #minetest |
| 19:14 |
|
IceDragon joined #minetest |
| 19:21 |
|
detrout joined #minetest |
| 19:25 |
|
rymiel joined #minetest |
| 19:27 |
|
ssieb joined #minetest |
| 19:52 |
|
appguru joined #minetest |
| 20:27 |
sfan5 |
IceDragon: https://github.com/minetest/minetest/blob/master/src/script/lua_api/l_item.cpp#L430-L443 |
| 20:28 |
|
Extex joined #minetest |
| 20:29 |
IceDragon |
From a glance the only culprit I could think of is the read_item, I should be able to find the function unless you already know where it is |
| 20:30 |
sfan5 |
probably in cpp_api/c_content.cpp |
| 20:31 |
|
nore joined #minetest |
| 20:32 |
IceDragon |
Yep that's where it was |
| 20:32 |
IceDragon |
Doesn't look like it would be that inefficient, unless I'm missing something |
| 20:34 |
IceDragon |
wait a second, the itemstack is being returned by value, which implies a copy |
| 20:34 |
sfan5 |
https://en.cppreference.com/w/cpp/language/copy_elision nah |
| 20:34 |
|
grouinos joined #minetest |
| 20:36 |
IceDragon |
I guess one way would be to just benchmark creating 4 million C++ itemstacks and see what it does |
| 20:48 |
IceDragon |
okay gonna try something different, changing the m_aliases to a regular map (from unordered_map) and running the benchmark again |
| 20:48 |
IceDragon |
I doubt it would make a difference |
| 20:49 |
sfan5 |
do you create empty item stacks or ones with name? |
| 20:49 |
sfan5 |
also you should profile this properly, for example with perf |
| 20:50 |
IceDragon |
ones with names |
| 20:50 |
IceDragon |
C++ isn't my strong point so don't expect much of me there P: |
| 21:04 |
|
IcyDragon joined #minetest |
| 21:05 |
IceDragon |
The struggles of crappy mobile internet |
| 21:16 |
IceDragon |
sfan5: How do I use perf, that aside I split my test into different types, empty item stacks complete in 1.7~ seconds, itemstrings take 11, tables take 8, tables with a blank name take 7, since my initial test was doing all four at once this gives an idea of the issue |
| 21:17 |
sfan5 |
I'd suggest this https://www.brendangregg.com/FlameGraphs/cpuflamegraphs.html |
| 21:18 |
sfan5 |
I tried this on the entire client yesterday: |
| 21:18 |
sfan5 |
perf record -F 400 -a --call-graph dwarf -- ./bin/minetest --go --worldname test |
| 21:18 |
sfan5 |
perf script | /tmp/FlameGraph/stackcollapse-perf.pl | /tmp/FlameGraph/flamegraph.pl > graph.svg |
| 21:18 |
sfan5 |
but that didn't work so well, what you want to do is minimize the testcase so only your benchmark code ever runs |
| 21:20 |
IceDragon |
yeah that's gonna be hard since it's in lua land right now |
| 21:20 |
IceDragon |
I could always stick it in a single mod under a clean world |
| 21:20 |
IceDragon |
since all it needs to do is create the item stacks |
| 21:23 |
sfan5 |
yeah maybe try that first |
| 22:00 |
|
Sven_vB joined #minetest |
| 22:31 |
IceDragon |
Okay sorry about the delay something work related came up |
| 22:31 |
IceDragon |
I have the perf.data so now I'm gonna read the flame graphs |
| 22:56 |
IceDragon |
Okay I've tested, empty stacks, itemstring, table and itemstack from another itemstack, all graphed up as svg |
| 23:09 |
erlehmann |
IceDragon, why are you creating 4 million itemstacks in the first place |
| 23:10 |
IceDragon |
I was testing a inventory serializer, in order to ensure that it performs well I ran it with an absurd number of times and noticed that it took way longer than expected |
| 23:10 |
IceDragon |
Basically: I had to hammer the code to get noticeable benchmark numbers |
| 23:11 |
IceDragon |
under normal circumstances I wouldn't care, but because it seemed to be taking a little too long I wanted to figure out why |
| 23:11 |
IceDragon |
If I can create 1 million tables under a second, I can't fathom why an ItemStack takes 10x that |
| 23:26 |
IceDragon |
https://share.druid.space/2021-09-16/minetest/itemstack-graphs/ |
| 23:26 |
IceDragon |
I uploaded the result from my perf tests |
| 23:32 |
sfan5 |
I wonder why time is spent on Toolcaps when there aren't any |
| 23:33 |
IceDragon |
Not sure if it's worth fixing since this is an extreme (test) case, but seems like there is room for optimization |
| 23:35 |
MTDiscord |
<Jordach> i think i have an idea why |
| 23:35 |
MTDiscord |
<Jordach> all tools correct fallback to the players hand definition of ":" |
| 23:35 |
MTDiscord |
<Jordach> in other words it's templating every itemstack |
| 23:36 |
MTDiscord |
<Jordach> to copy the default hand tool props |
| 23:37 |
MTDiscord |
<Jordach> this includes craftitems, tools with no defined properties, nodes |
| 23:37 |
sfan5 |
that would be stupid |
| 23:37 |
MTDiscord |
<Jordach> i mean this is minetest |
| 23:37 |
MTDiscord |
<Jordach> stupid seems to be the norm |
| 23:38 |
erlehmann |
second hand when |
| 23:41 |
|
Lunatrius` joined #minetest |
