dgilmore changed the topic of #fedora-riscv to: Fedora on RISC-V https://fedoraproject.org/wiki/Architectures/RISC-V || Logs: https://libera.irclog.whitequark.org/fedora-riscv || Alt Arch discussions are welcome in #fedora-alt-arches
tg has quit [Quit: tg]
tg has joined #fedora-riscv
davidlt has joined #fedora-riscv
davidlt has quit [Remote host closed the connection]
davidlt has joined #fedora-riscv
<rwmjones_> damn, ghc build (with itself) failed with
<rwmjones_> ===> Command failed with error code: -9
<rwmjones_> right at the final linking step!
<rwmjones_> I think plan B for me now is to look at the code to work out what the interpreter really requires
<rwmjones_> this comment seems relevant:
<davidlt> rwmjones_, spin another builds on the side you are doing it
<davidlt> Sadly in RISCV land the best option is still work on multiple paths in parallel
<rwmjones_> am going
<rwmjones_> am doing
<davidlt> rwmjones_, so the other options is to do cross-compile bootstrap from zero
<rwmjones_> I'm still looking into the code to try to understand why the heck the "internal-interpreter" flag isn't set
<rwmjones_> it's incredibly confusing
<davidlt> yeah, I recall looking into it
<davidlt> what would happen if you find that check, and just forced it to True?
<rwmjones_> it's a flag which is spread all over the code base
<rwmjones_> still investigating ..
<davidlt> rwmjones_, having a quick look and it seems these changes are relatively young, 3 years
<davidlt> This was never part of GHC8 IIUC
<davidlt> Thus I was living a happier life with GHC 8.X until I bootstrapped GHC 9 with GHC 8.
<davidlt> I wonder why rebuilding exactly the same compiler twice doesn't enable the internal interpreter
<rwmjones_> posted that q just in case
<davidlt> I mean, we have exact same version for the initial stage.
<rwmjones_> well i dont know what happens in the same compiler case b/c my first build failed
<rwmjones_> trying again now
<davidlt> Because this should be true: bootCross = (==) <$> ghcVersionStage Stage0 <*> ghcVersionStage Stage1
<rwmjones_> ok i see what you mean
<rwmjones_> idk, still looking ....
<davidlt> Yeah, just bumping and rebuilding
<davidlt> The 1st one doesn't incl. internal interp, but the second one should get it (I would assume).
<davidlt> As long as Fedora doesn't modify version somehow it should be true.
<davidlt> ghcVersion :: [Int] <- fmap read . splitOn "." <$> expr (ghcVersionStage stage)
<davidlt> Ah, sorry I should look at ghcVersionStage:
<davidlt> | Look up the value of a 'Setting' in @cfg/system.config@, tracking the
<davidlt> The Glorious Glasgow Haskell Compilation System, version 9.4.5
<davidlt> So Fedora aren't doing anything custom the versioning
n0nz3r0 has joined #fedora-riscv
<davidlt> I am launching a build on x64 to check that config file
<davidlt> ghc-version = 9.4.5
<davidlt> project-version = 9.4.5
<davidlt> rwmjones_, in the logs I have:
<davidlt> Building GHC version : 9.4.5
<davidlt> Bootstrapping using : /usr/bin/ghc
<davidlt> which is version : 9.4.5
<davidlt> Exact same version used
n0nz3r0 has quit [Ping timeout: 256 seconds]
<rwmjones_> FYI as there was a question about s390x before, on s390x:
<rwmjones_> [root@s390x-kvm-103 ~]# ghc --interactive
<rwmjones_> GHCi, version 9.4.5: https://www.haskell.org/ghc/ :? for help
<rwmjones_> ghci>
<rwmjones_> Leaving GHCi.
n0nz3r0 has joined #fedora-riscv
<rwmjones_> may have cracked this (with help from reddit): https://src.fedoraproject.org/rpms/ghc/pull-request/7
<rwmjones_> I'm testing a local build now
mangelajo has quit [Ping timeout: 268 seconds]
n0nz3r0 has quit [Ping timeout: 240 seconds]
n0nz3r0 has joined #fedora-riscv
<davidlt> Ah, did I miss it for GHC 9?
neil has quit [*.net *.split]
tibbs has quit [*.net *.split]
oaken-source has quit [*.net *.split]
skip77 has quit [*.net *.split]
kito-cheng has quit [*.net *.split]
djdelorie has quit [*.net *.split]
oaken-source has joined #fedora-riscv
skip77 has joined #fedora-riscv
kito-cheng has joined #fedora-riscv
skip77 has quit [Changing host]
skip77 has joined #fedora-riscv
tibbs has joined #fedora-riscv
neil has joined #fedora-riscv
<davidlt> which GHC version has it merged?
djdelorie has joined #fedora-riscv
<rwmjones_> davidlt: I think none, just master (ie. 9.9)
<davidlt> rwmjones_, Cabal stuff is probably in 9.9 too?
<davidlt> So basically this lands in 9.10, and it's gonna take quite some time until this can be removed.
<davidlt> rwmjones_, the previous patches on GHC were already merged
<rwmjones_> davidlt: I don't think PR#7 was merged yet?
<davidlt> All your previous PRs are merged
n0nz3r0 has quit [Remote host closed the connection]
<rwmjones_> davidlt: ok so all the PRs are merged now
<rwmjones_> the situation is that I'm still building ghc locally, to check that the interpreter problem is fixed
<rwmjones_> *if* it is fixed then I will close those other PRs
<rwmjones_> and rebuild the packages and hopefully pandoc should be rebuildable
<rwmjones_> this is still going to take ~ 48 hours I expect
<rwmjones_> and that's assuming nothing goes wrong
<rwmjones_> you might want to submit ghc (from rawhide) through koji at the same time, especially if it builds faster on milkv
<rwmjones_> btw fedora 40 branches is happening today I think
<davidlt> rwmjones_, as-is it's limited to -j1, it will not be faster on pioneer
<davidlt> f40 branching is tomorrow
<davidlt> Most likely US-time thus tomorrow just "starting" :)
<rwmjones_> ah, I didn't include the -j1 part of your patches .. I'm not sure that's necessary? seems to build ok in parallel here (except that time when it randomly failed ....)
<davidlt> I have a comment not to enable -j<ncpus> until 9.6 and LLVM 14.
<davidlt> I had some strange random fails without -j1.
<rwmjones_> that may need another PR in that case
<davidlt> I could have been fixed by now. We will see.
<davidlt> My commit message says retest once GHC 9.6 becomes default with LLVM 14.
<rwmjones_> ack
<davidlt> It could have been that vDSO issue (same as Rust)
<davidlt> That was a kernel fix.
<rwmjones_> I've kicked off another ghc build, this time under qemu with -j16
<davidlt> rwmjones_, ok
davidlt has quit [Ping timeout: 268 seconds]
fuwei has quit [Ping timeout: 255 seconds]
fuwei has joined #fedora-riscv
<rwmjones_> -j16 has almost caught up with the -j1 build I'm doing on the vf2