Time Nick Message 17:02 sfan5 rubenwardy: I think it'd be useful to limit the CDB update check to once every 30m or once during the runtime of the MT process 17:02 sfan5 right now just going in/out of a world will cause a recheck 17:06 rubenwardy Makes sense. It's currently cached as a global but that won't persist when the mainmenu env dies 17:08 sfan5 you could use core.get_once / core.set_once 17:08 rubenwardy that's not documented in menu_lua_api.txt? 17:09 sfan5 indeed 17:09 sfan5 but it exists :) 17:41 sfan5 it seems MT does not actually check the protocol version properly 17:41 sfan5 if you connect to a 0.4.x server it will try to do SRP and stuff and only abort because it doesn't understand the itemdef version 19:00 rubenwardy SDL2 should be documented as a dependency in doc/compiling 19:04 rubenwardy Debian 10 (end of life June 2024) only has SDL 2.09 in the repos. Minetest seems to require at least SDL 2.0.18 19:05 rubenwardy given the EOL is so close, we could stop supporting Debian 19:05 rubenwardy +10 19:09 sfan5 @paradust do you have time to PR fixes for the logging-related crash/issues? 19:14 MTDiscord for mingw? 19:15 sfan5 yes 19:15 sfan5 I remember you had some general improvments in mind that'd be better not just specifically on MinGW 19:18 MTDiscord I didn't really find a way to make mingw not crashy in some way. The options I identified are 1) replace all "thread_local" with a wrapper that uses a mutex and lookup table based on pthread_self(), or 2) same thing, but use a single thread-local initialized in Thread::run() to avoid the mutex 19:19 MTDiscord even after removing all the loggiing code, it was still crashing (not as frequently, but maybe 10% of the time) from other thread locals 19:19 sfan5 well that's unfortunate 19:20 sfan5 time to drop windows support? 19:23 MTDiscord maybe mingw support. What would be the downside of shipping an MSVC or ClangCl build? 19:23 rubenwardy for my own project, I gave up on MinGW due to threading issues as well and just ship MSVC 19:29 MTDiscord Irrlicht built with SDL doesnt accept input if you switch to second layout(e.g. ru_RU) on wayland btw. This doesnt happen on xorg 19:30 MTDiscord Also it doesnt accept input from some keys(e.g. OEM 3 aka `) on both xorg and wayland 19:49 [MTMatrix] It'd be great to see some sfence PRs merged, as we might be faced with a potential core dev that can reduce (and invert?) the PR stacking we've been seeing recently 21:30 grorp merging #14126, #14199 in 15 min 21:30 ShadowBot https://github.com/minetest/minetest/issues/14126 -- Method add_pos for object/player by sfence 21:30 ShadowBot https://github.com/minetest/minetest/issues/14199 -- Comply with base64 library license terms by appgurueu 21:49 grorp done 22:40 sfan5 @paradust there are some minor disadvantages to using MSVC but no dealbreakers 22:41 sfan5 C++ standards support is thankfully not an issue anymore 22:57 sfan5 1. coredevs who use linux can no longer produce builds, verify issues or help with the windows workflow 22:58 sfan5 2. getting our deps from vcpkg moves some control out of our hands e.g. lib versions 22:59 sfan5 and last time I checked they also patch some libraries weirdly. the ones I build are 100% upstream source code and in some cases I also disable unused features to save binary size. 23:00 sfan5 3. not sure if gdb can read msvc debug information. I have no expertise with msvc's debugger and neither does MT's '--debugger' option. 23:02 sfan5 4. I think there's more weird msvc workarounds in the code than weird mingw workarounds :) 23:03 sfan5 that's all I can think of, time for bed 23:04 rubenwardy Don't you need a Windows VM anyway to verify issues? 23:05 MTDiscord If you're thinking you may have to retire Windows official support due to lack of developers who are willing/able to work on it, maybe it's best to announce a call for volunteers soon and see if you can net any. 23:05 MTDiscord Wow, I didn't think this day would come 23:06 sfan5 '<+sfan5> time to drop windows support?' was a joke fyi 23:07 sfan5 rubenwardy: cross-compiling and using a random vm / grabbing some windows computer is easy; MSVC + relevant stuff is tedious to install 23:13 MTDiscord I don't think it makes sense to voluntarily drop Windows support, but if nobody is willing to actually support it then it's not really voluntary anymore. You can only choose whether to lie about it or not at that point. 23:29 MTDiscord It would be worth spending a little more time trying to figure out why SegmentHeap makes the thread_local problems so much worse. And coming up with a minimal example (a single .cpp file) that demonstrates a consistent crash. Maybe the mingw folks will take it seriously then. 23:30 MTDiscord Not sure we want to tie ourselves to MingW32 / 64 anyway, which seem to be in a bad shape 23:30 MTDiscord Maybe WSL has taken a big chunk of their potential user base?