IRC channel logs
2023-03-05.log
back to list of logs
<stikonas>fossy: do you remember why dist does not install into DESTDIR? <pder>stikonas: I believe you might need to do grep -h or grep --no-filename with that grep command using run*.sh <pder>Otherwise you will iterate over names like run.sh:bash-5.2.15 <pder>when you specify only one file grep doesnt prefix the results with the filename, but with multiple files it does <stikonas>ok, at least that old grep still has this option <pder>cool, thanks. how much disk space is actually used at the end of sysa? <stikonas>pder: you can insert bash command at the end and check <stikonas>I think the maximum usage is just after linux linking stage <stikonas>so if you add "bash" at the end of src_compile <stikonas>and then you could take a look at /proc/meminfo <stikonas>fossy: do we want to keep independent .env files for sysa and sysc? <stikonas>some variables are changing, so if we transfer it, we would have to source it, update some variables and reexport <stikonas>so not sure whether it's worth transferring it <rickmasters>stikonas: the new bash is raising new kernel bootstrap issues. <rickmasters>sitkonas: I'm looking forward to post-integration when the impact of these changes can be more easily assessed. <stikonas>gforce_de1977: unbz2 is missing in your run, did you update submodules? <stikonas>rickmasters: yeah, once it is integrated, it will be easier to make those changes <stikonas>rickmasters: do you have any details of what is going wrong with new bash? <stikonas>rickmasters: if we know what causes it, we could see if there is a version of bash between 2 and 5 that works <rickmasters>It's a few faults in configure resulting in different answers which lead to checksum differences. <stikonas>should be fixable by forcing the answer to configure tests <rickmasters>stikonas: the first is that divide by zero is not producing SIGFPE <rickmasters>stikonas: Page Fault after: checking for presence of POSIX-style sigsetjmp/siglongjmp... <stikonas>I guess once we start using autotools, we should expect reproducibility issues with different kernels :( <gforce_de1977>stikonas: just, updating the submodules fixes it - sorry for the noise, i closed the report in github <rickmasters>stikonas: Page Fault after: checking if signal handlers must be reinstalled when invoked... <stikonas>hmm, do you think it's possible to workaround those then? <stikonas>or do you want me to change something with bash then <stikonas>ok, let me know if you see that it's a lot of work... <stikonas>my initial goal was to get to as new autotools as possible <stikonas>but then I realized it's getting very hard to build binutils newer than 2.30 <stikonas>and 2.30 does not yet need automake 1.16 (which needed newer bash than 2.05b) <stikonas>gforce_de1977: yes, soon we won't need a pre-built kernel in live-bootstrap <stikonas>but we are still not quite yet there, though getting fairly close <rickmasters>And the reason for doing it in sysa was to get bash interactivity as soon as possible? <stikonas>rickmasters: no, see that automake stuff <stikonas>I thought we'll be able to update more stuff <stikonas>but if we just want interactivity, I could rebuild bash 2.05b too with musl <stikonas>your issues might be there too (caused by differences in autotools on different kernels) <stikonas>gforce_de1977: by the way, you need 32-bit kernel, 64-bit kernel won't work <stikonas>you'll be able to run sysa on 64-bit kernel but the transition with kexec won't work <rickmasters>stikonas: forcing the answer to configure tests sound easier than kernel debugging. Any hints on how I can try that? <stikonas>you basically just pass it to ./configure script <stikonas>I suppose ./configure system environment autodetection is a bit at odds with reproducibility in general <stikonas>we should be fine with libc and userspace tests but once they start testing kernel features we are in trouble <muurkha>the reason the TH1520 is interesting is that it's a fully open-source RISC-V implementation which can evidently deliver fairly reasonable performance when implemented in a modern process (2.4 GHz in 12nm in this case) <gforce_de1977>stikonas: is there a known-to-work-32bit-kernel downloadable somewhere? <stikonas>probably slightly outdated, so just use it for live-bootstrap and not for production <rickmasters>stikonas: I was able to force matching answers for the three obvious configure problems. <rickmasters>stikonas: unfortunately, there is still a checksum mismatch so I'll have to look closer. <rickmasters>stikonas: I've still got an issue. I'm wondering if you could at least hold off on further changes to sysa until I can resolve this. <stikonas[m]>I'll just push non package change (pass environment variables via file).That wouldn't affect fiwix <rickmasters>stikonas: Thanks. I'm hoping I can work through this soon. <stikonas[m]>Though we updated binutils from 2.14 to 2.30. That will help us longer turn <rickmasters>stikonas: don't worry about it. I'll try to review your PRs in advance until integration. I have to step away for a while. Back later. <stikonas>gforce_de1977: still stuck with mkdir call? <stikonas>make sure that you don't have unnecessary files in live-bootstrap directory <stikonas>one thing that might go wrong is that those files will make initramfs image too big