IRC channel logs

<oriansj>the only problem i predict with this change is it break's mescc' !0xDF !0xFC strings (which i don't understand when we have 'DF FC' raw strings in M1)

<oriansj>or we special case !## to work for -127 to 255 and leave it to the user to know what they are doing.

<stikonas>oriansj: I've switched mescc to raw strings

<stikonas>though that is still pending for the next release

<stikonas>oriansj: but the next mes/mescc release would need new stage0-posix release anyway (at least on 64-bits)

<stikonas>oriansj, janneke: so perhaps stage0-posix release should be made at the same time as the next mes release

<oriansj>well that solves that concern.

<oriansj>and I am good with that.

<stikonas> (https://gitlab.com/janneke/mes/-/commit/061d7f93800e39b9927f748c9d1c17eaffb7e106#15d34292635b3d1487809ba9d4393e5f6ca36149_143_148)

<stikonas>it's still on wip branch for now

<stikonas>I need to create a gnu style commit message and add copyright statements but I just returned back from holiday, so haven't done that yet

<oriansj>ultimately it will be up to builder-hex0 and mescc to dictate what they want as standard M1 behavior.

<theruran>on my new Void Linux-musl system, and `sh kaem.amd64` fails with `kaem.amd64: 30: ./bootstrap-seeds/POSIX/AMD64/kaem-optional-seed: Permission denied`

<theruran>kaem-0 and bin/kaem: not found

<theruran>cleanup.sh also failed. couldn't find pushd???

<theruran>feel like I am missing some essential utility

<stikonas>theruran: are you running with bash or some other shell?

<stikonas>pushd is bashism

<theruran>fish

<stikonas>might be the cause

<stikonas>and I recommend running kame scripts with kaem or kaem-optional-seed

<stikonas>they might not be fully compatible with posix shells

<theruran>I ran it wish fish before. and cleanup.sh starts with /usr/bin/env bash which seems to work

<stikonas>oh ok, that's fine then

<theruran>what's the difference between kaem and kaem-optional-seed?

<stikonas>kaem is a C program

<stikonas>with more features

<stikonas>kaem-optional-seed is handwritten hex0 code

<stikonas>which only knows how to run other programs (with command line arguments) and 2 types of comments

<theruran>kaem-optional-seed is the one I get permission denied

<stikonas>kaem has e.g. aliases, conditionals, environmental variables

<theruran>because it's not a PIE?

<stikonas>it's not

<stikonas>it's the only binary in stage0-posix-amd64 that is not PIE

<theruran>OK

<stikonas>you can create PIE version

<stikonas>it would be a few bytes bigger

<stikonas>perhaps it makes sense to port

<stikonas>theruran: e.g. lines like https://github.com/oriansj/stage0-posix-amd64/blob/master/GAS/kaem-minimal.S#L121

<stikonas>mov eax, OFFSET hard

<stikonas>have to be replaced with lea eax, [rip+hard]

<stikonas>or possibly lea rax, [rip+hard] (whichever has smaller x86_64 encoded size)

<stikonas>theruran: so you system can only run PIE binaries?

<stikonas>I suspect some later stages of bootstrap might break...

<theruran>I guess so. seems to be the standard nowadays

<stikonas>not sure how well old gnu software supports requiring PIE

<theruran>hmmm

<stikonas>but perhaps you can pipeclean it to work

<stikonas>nobody tried running bootstrap on hardened system that only supports PIE

<theruran>ACTION :)

<stikonas>anyway, in stage0-posix and mes stages you'll only have to fix kaem-optional-seed

<stikonas>no idea what will happen with old tcc...

<oriansj>well old tcc didn't support making Position Independent Executables

<oriansj>mostly because trying to doing that on x86 sucks

<oriansj>(you literally have to do call 0; pop rax and then you can do mov eax, [eax+offset])

<oriansj>^pop rax^pop eax^

<stikonas>well, in any case we don't have a working tcc-0.9.26 bootstrap on amd64

<stikonas>only on x86...

<stikonas>and I'm right now running a test with riscv64 with wip-riscv branches from janneke and ekaitz

<stikonas>already running for 2h...

<stikonas>(qemu user emulation is being slow...)

<stikonas>janneke: tcc 0.9.26 is probably not too far from building

<stikonas>compiler (mescc) seemed to be happy on wip-riscv branch but linking stage errored out with

<stikonas>"Target label expr_const64 is not valid"

<stikonas>probably a bit more patching is needed

<stikonas>before HAVE_LONG_LONG is available

<janneke>stikonas: very good news; patching of mescc or of tinycc (or both?)

<janneke>oriansj: very nice

<stikonas[m]>janneke: I think just tinycc

<janneke>nice

<stikonas[m]>Well, the fact that it compiled but failed to link means we called some function that was #ifdefed not to exist

IRC channel logs

2023-06-06.log