IRC channel logs
2025-03-24.log
back to list of logs
<oriansj>the reason has to do with M2libc compatibility with M0 and the hex2 written in hex1. <oriansj>M2libc is free to use all M2-Planet features except the bootstrap files which are quite restricted (and slow) <oriansj>and mescc-tools doesn't leverage the performance advantages the full M2libc provides until the second round of bootstrapping. <stikonas>I was mostly saying it's also other C features that are restricted <oriansj>but kaem could be built after M1 and hex2's C versions and thus could avoid the --bootstrap-mode entirely <oriansj>but I worry about nondeterministic behavior in kaem <LaurentH>cosinusoidally: Cool! It doesn't sound so bad, since __floatundixf is probably never called? Also, for the __va_start function that gcc is also reporting as missing, did you use tcc/lib/va_list.c? <LaurentH>Also, I managed to get tcc-by-pnut -ar working and made libc-mini.a by following mes-0.27/pass1.kaem in live-boostrap. I'm still reading live-boostrap's scripts, but it seems we may be able to build mes's libc using tcc-by-pnut and then recompile tcc with that libc? <stikonas>yeah, kaem is indeed built without --bootstrap-mode <stikonas>and perhaps we don't need to add nondeterministic behavious there given that we sped up meslibc build significantly using unified-libc file <fossy>stikonas: i think maybe not needed.... <fossy>LaurentH: that libc path sounds doable! <fossy>if you can get all the way to libc+tcc in mes-0.27 <fossy>(that is, literally libc+tcc.a) <matrix_bridge><cosinusoidally> LaurentH: yep I also needed to compile va_list.c (which tcc-by-pnut compiled fine) <ekaitz>janneke: i wrote a super simple benchmark setup in the simple.make file of mes, should we include that in the repo? <ekaitz>i'll send you the patch an let you decide <ekaitz>janneke: good! my approach has been the simplest possible <ekaitz>oh also mes-gcc is like twice as fast as mes-m2 <ekaitz>let me try to remove optimizations <ekaitz>-O0 makes one of the runs a 20% slower <stikonas>janneke: by the way, we'll have to sync more M1 defines from M2libc to meslibc... <stikonas>though these fill be for the future release of M2-Planet 1.13... <janneke>stikonas: okay, thanks for the heads-up <stikonas>should I send a patch on top of the previous one? <ekaitz>janneke: i saw your email, isn't the guix.git the place were I've put the dependency?