IRC channel logs
2026-03-26.log
back to list of logs
<daviid>gabber: you might wana ask in #spritely as well (though most of #spritely members are here as well ... but they are developing hoot ... <gabber>lechner: i am not sure guile can self-host, actually? MES can, and through TCC we can build ancient GCC and with a modern GCC we build guile, but i may very well be mistaken <dsmith>sneek, later tell ekaitz Whoo!! Congrats! <lechner>gabber / thanks, but i didn't actually ask about self-hosting. I just noticed several different "stages" and would like to find out what they are <rlb>lechner: if you mean guile itself, I don't know all the details well, but the steps, when bootstrapping from scratch may be (roughly), "first build an interpreter via C", then build the compiler using that interpreter, then build again using that compiler, etc. <sneek>Welcome back ekaitz, you have 1 message! <sneek>ekaitz, dsmith says: Whoo!! Congrats! <ekaitz>but yes, it was like that before too <rlb>The build/check was fine, but I may have to redo it... I forgot the --prefix, and I'm not sure you can change that without a make clean, or similar. <rlb>Or rather, I know a make clean will work, and iirc just a re ./configure and a make doesn't. <ekaitz>i'm also re-running with the suggestion of removing the fence that old told me <rlb>Oh, OK, should I continue testing this one too? (That's fine, just checking.) <ekaitz>the change old suggested is a small thing <graywolf>Is it expected that the ~/.cache/guile directory does not respect the $HOME variable? Or is that a bug? <graywolf>ooooh, (@ (foo) bar) causes (foo) to be loaded even when inside a (when #f ...), I did not expect that <ekaitz>old: can it be that removing the fences it deadlocks? <rlb>I tried just reconfiguring and that failed in the guile-fibers build as I'd expected, so I'll start over. Be another 8+ hrs or so. <rlb>Though now that I think of it, I suppose I probably could have tried to save my .go files... <rlb>I suspect that at least one thing that's causing reconfiguring a --prefix to produce a broken install is the pkg-config file, i.e. my current guess is that it's missing a dep somewhere so that it's rebuilt whenever the configure --prefix changes. <rlb>s/so that it's rebuilt/that will rebuild it/ <old>ekaitz: I don't see how that could deadlock .. maybe it loop indefinitly in that the operation always fail ..\ <old>anyway, it's an optimization that we could revisit as some point <old>if the initial fix worked, let's go with it <ekaitz>old: i run the tests several times and they got stuck a couple of times <ekaitz>maybe I didn't run the tests with the fence enough times to find that kind of issues <ekaitz>btw old you are the guy from blue? <old>the fencer is slower so it might be harder to reproduce <ekaitz>oh! so we actually know each other! <ekaitz>not in person but we have talked via email in the past <old>but yeah: Olivier Dion -> old <old>even tho, I am young, apparently <old>so, to me the fix is good to merge btw <old>forget about the removal of fences for now <old>I guess i'll let rlb merge whenever he gots the result back <old>I am a bit suprise that it does not work tho. The link I post show the implementation of CAS <old>would be interesting to see the generate instructions from the JIT an comparse against the "official" way of doing a CAS <ekaitz>yeah, i took the implementation from there + the arm backend of the jit