IRC channel logs
2026-04-01.log
back to list of logs
<daviid>tohoyn i actually on't get it, sometimes it works perfectly, just tried now, launch the app and 'straight' use the arrow keys to rotate the sphere, and everything is fine, no warnings, no segfault - i just tried a couple of times now, and here, debian testing, everything is fine - fwiw, I did see a libegl patch came in debian after we started to talk about this example, now it works perfectly fine (again), it seems so, here at least ... <rlb>jcowan old: yes, the gc'ed data should be eligible for conservative collection after leaving the function; I was just wondering if that would still be a reasonable "typical path" for C code with whippet (as compared to whatever else like dnywind/defer/whatever), and I'd also tried to keep the stack requirements "small", which I think we have up to now (though maybe larger than some of the existing limits), though the utf-8 limits would <rlb>be fairly easy to change. <rlb>I would imagine the "extension library" role and/or if we care about working with "thousands (and thousands) of threads" might affect our thresholds, but not sure what our better defined requirements are, if any. <rlb>I'd also assumed that say 512-1024 might cover a large number of the common cases, e.g. symbols, filenames, functions, etc. <rlb>Overall, no strong opinion. <old>rlb: is this documented in the utf-8 branch? <old>I'll take note when looking the commits <rlb>old: the intention is for the commits/comments/messages to tell a reasonable story, though some still need work, in part, I assume, based on eventual review-related questions. Though the series also evolved as my understanding improved, so at the moment, there's variation, depending on when bits were implemented. Those can certainly be reworked if we decide we'd like to. <rlb>Something like the "Convert char-set->string to UTF-8 via new begin/finish API" early in the series might provide an example. Because the series tries to keeps the tests working throughout, the transition is taken in stages. <rlb>(Though I think I may need to recheck the tests in a few places after some rebasing.)