IRC channel logs

<matrix_bridge><gtker> oriansj: If raw bytes wrapped in '' should be used for u8s, and % for u32s what would be the correct way of outputting an array of u16s?

<matrix_bridge><Andrius Štikonas> Hmm, I guess we don't have a way that works for both normal arches and word based...

<matrix_bridge><gtker> stikonas: It's not required for the current bootstrap target (tcc_1_7) but it might become relevant at some point

<ekaitz>stikonas: m2-planet release with switch/case support?

<stikonas>ekaitz: let me check changelog

<stikonas>yeah, last release should support switch/case

<ekaitz>last is 1.11.0?

<ekaitz>that's from 2023

<matrix_bridge><Andrius Štikonas> hmm

<ekaitz>i'm looking at github

<matrix_bridge><Andrius Štikonas> I think so

<ekaitz> https://github.com/oriansj/M2-Planet

<stikonas> https://github.com/oriansj/M2-Planet/blob/0236ee4f8533258de7442246e0fba2d7b5c42fcf/cc.c#L53

<stikonas>we'll need to bump it soon

<stikonas>well, there are lots of new things in git master

<ekaitz>i'm using it, so if you bump it, please let me know!

<ekaitz>hmm looks like 1.11.0 doesn't support it because it's failing

<ekaitz>ACTION is sad

<matrix_bridge><gtker> stikonas, ekaitz: I believe https://github.com/oriansj/M2-Planet/commit/5566930976d23f0c6efeb03cf6cc19cf04cb61c1 added switch statements on Nov 10, while 1.11.0 was released on Nov 01. It seems we forgot to add switch statements to the "Current" changelog header

<ekaitz>oh ok so i need a newer version

<matrix_bridge><Andrius Štikonas> hmm, we should do a new release then...

<matrix_bridge><gtker> We don't have a release with switches, but the current master does

<stikonas>hmm, should we make one after I merge current PR?

<matrix_bridge><gtker> stikonas: I don't have anything urgent to get in before the release, so we could do one now

<stikonas>gtker: do you think it's reasonably stable now?

<matrix_bridge><gtker> Maybe update the changelog with switches?

<stikonas>yeah...

<stikonas>well, need to merge your PR too

<stikonas>and do a bit more testing...

<matrix_bridge><gtker> I can do it in my PR

<stikonas>see if live-bootstrap works at least on x86

<stikonas>ok, thanks. Please do so

<matrix_bridge><gtker> stikonas: PR updated

<stikonas>thanks

<stikonas>hmm, interesting, we did have stage0-posix release in Sep 2024

<stikonas>so that one includes new M2-Planet

<stikonas>but I guess there was no standalone release

<stikonas>gtker: do we not support size == 2?

<stikonas>oh yes, there is a comment

<ekaitz>hm! i'm trying it and it's not working

<ekaitz>is there any obvious limitation in the switch statement i'm missing?

<matrix_bridge><Andrius Štikonas> Hmm, what is your code?

<ekaitz> https://paste.debian.net/1362916/

<matrix_bridge><Andrius Štikonas> This is testcase that we have https://github.com/oriansj/M2-Planet/blob/master/test/test0031/switch.c

<ekaitz>my code is simpler than that testcase

<matrix_bridge><Andrius Štikonas> Yeah...

<matrix_bridge><Andrius Štikonas> So what is going wrong?

<matrix_bridge><Andrius Štikonas> Goes to the wrong branch?

<ekaitz>no, it fails

<ekaitz>the build

<matrix_bridge><Andrius Štikonas> Hmm

<ekaitz>but maybe it's just me fighting against the buildsystem

<matrix_bridge><Andrius Štikonas> Any useful error?

<ekaitz>not really

<ekaitz>maybe it's the mes scaffolding

<ekaitz>it's failing in kaem --strict -f kaem.run

<ekaitz>src/reader.c:460:switch is not a defined symbol

<ekaitz>hmmm

<matrix_bridge><Andrius Štikonas> Hmm

<matrix_bridge><Andrius Štikonas> And are you sure it is new m2 planet?

<ekaitz>i'm not sure if this is giving me the proper m2

<ekaitz>heh

<matrix_bridge><gtker> Isn't it because of `switch (c)

<matrix_bridge>{

<matrix_bridge>- n = reader_read_hex ();

<matrix_bridge>- c = n->value;

<matrix_bridge>- case '\\':`?

<matrix_bridge><gtker> Assignments inside the switch?

<ekaitz>isn't it legal to make assignments inside a switch?

<ekaitz>ah! haha! it's a commit body, those - lines mean the line is removed

<ekaitz>in gcc it works

<matrix_bridge><gtker> Ah yeah, nvm lol

<matrix_bridge><gtker> Can we switch on char literals?

<ekaitz>hmmm

<ekaitz>it's not working yet... and I refreshed the commit with the last one you did

<ekaitz>latest*

<ekaitz>hmmm

<ekaitz>i'm surely doing something wrong

<matrix_bridge><Andrius Štikonas> Hmm, we should sort this out before release...

<ekaitz>it's probably the guix packaging here that is wrong

<matrix_bridge><Andrius Štikonas> Your error suggests that it is switch statement itself that is unrecognised

<ekaitz>yes

<ekaitz>and i think guix is using the M2 of a previous environment and not this

<ekaitz>and i'm very upset

<ekaitz>:)

<stikonas>anyway, let us know if you figure this out

<stikonas>if we make a new release, bettter fix this

<stikonas>(if that's not a guix issue)

<ekaitz>i'm pretty sure it's guix but i don't know why

<ekaitz>do i need other mescc-tools if I change m2-planet?

<stikonas>ekaitz: probably not

<stikonas>though we had some breaking change, but it might be fine

<stikonas>well, the pointer arithmetic now works for += and -= operators

<stikonas>(but not for + or -)

<ekaitz>stikonas: scaffold/main.M1:7 :Received invalid other; xor_eax,eax

<ekaitz>but I don't have the switch issue anymore

<stikonas>oh, that's M2libc

<stikonas>I think you need newer defines

<stikonas>hmm, now how do they get added to mes

<stikonas>I guess mes keeps its own copy

<stikonas>it's these two commits: https://github.com/oriansj/M2libc/commit/52965694d58d846f94eaac3a897647c94e4959f4 https://github.com/oriansj/M2libc/commit/cbf165034fbd90b2f760185774cbcb17b8a27435

<ekaitz>oh shit

<stikonas>it's in ./lib/m2/x86/x86_defs.M1

<stikonas>should be easy to sync...

<stikonas>but yeah, that's something I forgot about

<ekaitz>just clone and overwrite

<stikonas>it's a bit annoying

<stikonas>cause any codegen changes in M2-Planet that need new defines

<stikonas>are certain to break mes bootstrap

<stikonas>well, "fix" is easy, just need to add those to mes

<stikonas>but then need to wait for a new release of mes...

<ekaitz>heh

<ekaitz>i don't know why mes doesn't get them from M2lib

<stikonas>hmm, maybe janneke can answer that...

<stikonas>well, it's fine for risc-v

<stikonas>since that list of defines is now basically complete

<ekaitz>and what are all these: DEFINE RETURN C3

<ekaitz>those are removed in M2lib

<stikonas>return opcode

<stikonas>maybe it still has ret?

<stikonas>oh, those are old style defines

<stikonas>before M2-Planet switched to newer style (more similar to GAS) defines

<stikonas>I guess those can go away soon from mes

<stikonas>especially once mes depends on newer features of M2-Planet

<ekaitz>oh

<ekaitz>i just used the new files

<ekaitz>and it didn't complain because of it

<ekaitz>src/reader.c:462:src/reader.c:462:ERROR in process_switch

<ekaitz>MISSING }

<ekaitz>Subprocess error 1

<ekaitz>ABORTING HARD

<ekaitz>but it complained about that instead

<stikonas>ok, that's the error from new M2-Planet

<stikonas>hmm, I guess MISSING } means it saw something unexpected?

<stikonas>ekaitz: that's with the same switch statement from your pastebin?

<ekaitz>yes

<ekaitz>maybe the \\?

<ekaitz> https://paste.debian.net/1362935/

<ekaitz>that's the switch

<stikonas>ekaitz: yeah, that's chars...

<stikonas>this also fails https://paste.debian.net/1362936/

<stikonas>with the same error

<stikonas>or maybe something else there

<stikonas>still checking

<stikonas>but we have a simpler reproducer

<ekaitz>hehe

<ekaitz>i have a magnet for these things

<stikonas>ekaitz: I think switch can accept char but not case

<stikonas>case seems to need int

<ekaitz>okay so if I write a number instead it should work?

<stikonas>I think so

<stikonas>you might also need to add default

<ekaitz>can we improve M2planet to support chars?

<ekaitz>does it support enums?

<stikonas>well I'm looking now

<stikonas>it seems to me that it expects only 1 token there https://github.com/oriansj/M2-Planet/blob/6341097a328b7a2cc2a0325b54fb335072fc9772/cc_core.c#L2783

<stikonas>'a' I guess are 3 tokens

<stikonas>gtker: can we used constant_expression here?

<stikonas>ekaitz: as for enums, yes, new one has support for enums

<stikonas>should be fairly good

<ekaitz>that's great

<stikonas>oh, buti not sure

<stikonas>if they work together with switch/case

<stikonas>so that might not help you...

<stikonas>we should try to fix it...

<ekaitz>yeah...that would make my work in mes way easier

<stikonas>let me open a bug...

<ekaitz>oh thanks

<stikonas> https://github.com/oriansj/M2-Planet/issues/99

<ekaitz>how hard do you think it is to add chars there?

<stikonas>should be doable I think

<stikonas>gtker might do if faster than me though

<ekaitz>hehe

<ekaitz>i tried to read the code and it doesn't look impossible

<ekaitz>but i need some concentration

<stikonas>but in the meantime I guess you could do something like case 0x61: // 'a'

<ekaitz>yep

<ekaitz>it's working with the numbers

<ekaitz>so... not bad

<stikonas>well, I have a hacky commit that goes most of the way

<stikonas>and it also breaks some stuff :(

<stikonas>needs a space between case expression and :

<stikonas> https://paste.debian.net/1362945/

<ekaitz>hm!

<ekaitz>it makes sense

<stikonas>oh, that's because of some label hacks in M2-Planet tokenizer...

<stikonas>it thinks 1: is a label

<stikonas>and converts it to :1

<ekaitz>oh no

<stikonas>oh I guess that was previously dealt with if(':' == global_token->s[0])

<stikonas>so needs more thought on how to cleanly do it...

<stikonas>well, I might not have more time today

<stikonas>but maybe gtker can take over this patch :)

<ekaitz>thanks for the time :)

<oriansj>stikonas: just a thought but $ isn't used by word based architectures yet; what if we do a little shift to give word based architectures a standard 16bit symbol (as the byte based architectures are using @0x1234)

<stikonas>$ is used

<stikonas>I think all chars are used

<stikonas>$ is used for jumps

<oriansj>yes but not in M1

<stikonas>oh, maybe not

<oriansj>all of the chars can be dual use

<oriansj>one meaning in M1 (for immediates) and one meaning in hex2 (for jumps and calls)

<stikonas>indeed, $ does not seem to be used

<stikonas>but we don't want to implement that in M0...

<oriansj>you don't have to implement it in M0

<stikonas>yeah, just need to guard this feature

<stikonas>behind maybe_bootstrap_error

<oriansj>or we just take a day or two to backport it

IRC channel logs

2025-03-13.log