Discussion
git pretending it doesn't understand how to automatically merge changes in which one side added one unique line and the other added a unique line at the same spot of a makefile is:
(1) real fucking cute, linus
(2) literally the entire fucking reason to enforce tyrannical line-based syntax for the past half fucking century
go to hell, linus
i suspect this is a result of adversarially fucking with a .gitattributes file to make kconfig and makefile diffs exit nonzero for empty difference regions. i could automate it but you do actually learn the mechanics of how people break builds on purpose if you wallow in the hate. and linus has defeated me utterly before. so i'm not rushing through this encounter
mergiraf somehow saved me from a lot of shit so props for making good software that solves a problem even if it's absolutely the wrong problem to solve and just makes it easier for people to accept the fascist scheme where they break your build until you cry and let them backdoor you imho
i do genuinely respect that it can succeed at all here even if i specifically decided not to go that route long ago before i had words to explain why
@hipsterelectron have you looked at other models like patch based? I’m curious about your thoughts here
@RosaCtrl i think state based without any fucking merkle tree bullshit is the way to go and the implementation should take the form of a filesystem which supports transactional consistency guarantees to enable snapshots to have any meaning at all (this is btw my answer to "can recursive globbing ever have a consistent meaning", which is that the lack of any consistency (NOT global!!!!! that's the fucking billion-dollar mistake!!!!) is a fundamental design flaw in the entire field of filesystem design
@RosaCtrl in databases transactional consistency is widely accepted without requiring explanation by the mid-90s. the guy who invented ssh has a fucking fantastic master's thesis (the earlier one from his scandinavian university is MUCH better than the followup) with some great ideas and basically says "yeah thanks to memory being cheap now (90s) we can just not globally lock by always working on a scratch copy and swapping out asynchronously", which is literally my only idea for i/o perf
@RosaCtrl meanwhile OS designers are like "because people sometimes watch video streams and copying data is slow and the cpu has no support for copying asynchronously, we have no choice but to share memory fucking everywhere all the time" and CPU designers proceeded to go "oh, you want to copy memory in fixed-size power-of-two chunks and you want it to flush the entire pipeline for normal instructions and you want it to be limited to very expensive top-of-the-line hardware" and invented x86 SIMD and ARM NEON and ARM SVE and acted like copying memory isn't the single fucking operation the CPU caching model can absolutely do asynchronously, as if the entire fucking problem that breaks their LRU caching model is not this fucking incredible lack of a separate async memory copy capability
Install bonfire.cafe
Get the full app experience
this is a result of trying to update the 6.5 gcc lto patches (which do not successfully apply to 6.5) to 7.2. the kconfig is now complaining it can't possibly merge the intentional noise they added in the clang lto kconfig with the completely independent gcc lto kconfig. because linus, so magnanimous, renamed ARCH_SUPPORTS_CFI_CLANG, to merely ARCH_SUPPORTS_CFI. yet again, the inferred state at which their paths diverged irreparably is an empty string