IRC channel logs
2024-01-04.log
back to list of logs
<rickmasters>stikonas: i'd increase __FILEDES_MAX to much higher, like 512 <stikonas>rickmasters: oh thanks, I messed that I left it commented out <stikonas>instead of showing error message, it now hangs <stikonas>now both are showing a different error message <stikonas>didn't have much time today, but still we made some progress <stikonas>getting something like (unhandled exception: unbound-variable: (make-record-type) now <stikonas>in any case, even if we want full POSIX kernel at this stage <stikonas>it's probably easier to first rely on UEFI <stikonas>and then if we want, we port each syscall to what real kernel should do <stikonas>oh, that mes error is due to missing symlinks (same workaround is needed as live-bootstrap does, just copy files) <stikonas>though of course there is a next issue :( <stikonas>ok, looks like for some reason mes/module/mes/getopt-long.scm gets autocreated with zero size <stikonas>and then shadows module/mes/getopt-long.scm file <stikonas>moving latter file to the former location actually makes mes --version work, though this is just a workaround for now <stikonas>hmm, perhaps sys_access creates empty file on UEFI... <stikonas>yaeh, we are jus returning 0 there right now <Googulator>How do we currently deal with psyntax-pp in Mes? <matrix_bridge><Andrius Štikonas> Googulator: we unlink it in pass1.kaem mes build script <matrix_bridge><Andrius Štikonas> Though in builder-hex0 unlink is not implemented <Googulator>maybe we should do rename-then-delete in that case <Googulator>although its non-use is verified by bwrap working <matrix_bridge><Andrius Štikonas> Well rename is also not implemented in builder-hex0... <Googulator>I'm hoping to recover the pre-0.26 history this way. <matrix_bridge><Andrius Štikonas> Probably not, people had trouble finding old ghc versions... <snuik>oriansj was in #bootstrappable one day and 19 hours ago, saying: which would be surprising but certainly a good reason for them to change <oriansj>guess it couldn't handle stikonas' great UEFI work. <Mikaku>rickmasters: I'm reading your PR66 <stikonas>is this the last big piece of upstreaming? <rickmasters>stikonas: yes, there is one more small PR from Googulator for PAE and then we'll be able to pull directly from Mikaku <pder>Will the pregenerated files in nyacc issue go away with mes 0.26 now that it has module support? <stikonas>though you could try that other long tcc chain from cosinusoidally <stikonas>pder: it turns out that regeneration scripts require psyntax.pp from mes <stikonas>live-bootstrap rm's those files right now from mes <oriansj>nyacc is a bit of a problem in terms of performance and bootstrapping <stikonas>so for now eiter nyacc or mes is pregenerated <stikonas>well, performance is probably due to both mes and nyacc <stikonas>well, I guess if it was just performance, then it's not super big deal <stikonas>it's not like builder-hex0 or stage0-uefi are optimized for speed... <oriansj>well mes.c can always become a better scheme interpreter in regards to performance and features <pder>psyntax.pp can not be regenerated using mes? <stikonas>pder: well, somebody has to figure it out <stikonas>it was done for guile, but it was non-trivial effort <oriansj>pder: we can regenerate it from guile which was properly bootstrapped but that is long after we stop using mes.c <stikonas>it's probably taken from an older guile... <stikonas>I wonder how far we are from using M2-Planet directly instead of nyacc... <stikonas>10% of effort maybe has taken us 90% there, but last bits are probably more complicatedf <oriansj>well if you remove the need to do bit-slices from tcc (which just requires tcc to output assembly rather than binaries); then M2-Planet would be able to directly compile TCC <stikonas>there are always various small issues, e.g. tcc assumes 32-bit integers, etc... <stikonas>M2-Planet is not as good with operator precedence... <oriansj>potentially some C preprocessor bits but M2-Mesoplanet can be extended to handle those <oriansj>It complies with the C standard in regards to operator precedence but it doesn't yet short-circuit <oriansj>that is why there is that function call pattern up the operators exists in M2-Planet <oriansj>rather than just a flat if/else block in a single function <stikonas>yeah, it's not returning the same results as gcc <stikonas>I just tried 1 + 2 * 3 and gcc returns 7, M2-Planet 9 <stikonas>since M2-Planet doesn't look forward that much <stikonas>I guess to fix that you need abstract syntax tree <oriansj>probably just need to make + and * a different priority <oriansj>which is why ( works), it is a higher priority than +/*/etc <rickmasters>stikonas: fun fact: time from start of stage0-posix to post-mes is 15:28 for linux, 14:51 for builder-hex0 <rickmasters>lack of memory protection and faking syscalls has speed advantages <stikonas>I guess it's mostly loading file system that is slow <rickmasters>Mikaku: if you're still up I think its pretty late your location so please don't feel rushed to process PR until tomorrow <snuik>samplet was in #guix-risc-v one month and 19 days ago, saying: Hopefully not, but my first naive attempt at updating Mes there gave me errors, so I was checking if I could leapfrog having to debug them. :) <mihi>snuik, and in #bootstrappable?