IRC channel logs

<stikonas>oriansj: I've got simple global struct to work, so mix of e.g. integers and chars should be fine, stuff like (&a)->y seems to work too but right now I'm still having problems with arrays inside global structs...

<stikonas>do you think we can ignore them for the time being?

<stikonas>hmm, for some reason I think I get an extra dereference there...

<stikonas>hmm, that's probably something that might be fixable in dot processing function

<stikonas>after all . and [] kind of "cancel" each other in some sense

<stikonas>ok, adding if(match("[", global_token->s)) return; seems to fix it...

<muurkha>stikonas: how much complexity in the compiler was it to get the global struct to work?

<stikonas>muurkha: cc_core.c | 98 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++----

<stikonas>(that's not counting tests, etc...)

<stikonas>I'll see if I can get local structs to work too

<stikonas>probably similar complexity

<stikonas>in general not too complex

<stikonas>took me maybe an hour...

<stikonas>muurkha: and I think I've mostly got local structs to work too

<stikonas>well, at least on amd64

<stikonas>(might need adjustment for arches where stack goes the other direction

<stikonas>and I might have also found a bug in M1...

<muurkha>cool

<stikonas>on riscv it fails to convert RD_A0 RS1_FP !-0 ADDI

<stikonas>in particular leaves !-0

<stikonas>hmm, I wonder if M1 should be able to deal with !-0 or M2-Planet should not spit "-0"...

<stikonas>oriansj: what do you think?

<stikonas>M1 fix is not too bad https://github.com/oriansj/mescc-tools/pull/38

<oriansj>stikonas: -0 is just 0; unless you are telling me that riscv did something completely insane

<oriansj>M2-Planet should be smart enough to know not to spit out -0; in fact M2libc wouldn't ever produce that output if one did: int2str(x, 10, TRUE);

<oriansj>the only way you could get M2-Planet to output -0 is if you did: emit_out("RD_A0 RS1_FP !-"); which ment you were putting the incorrect offset in the tracking structures

<stikonas[m]>Hmm, yeah, ignore that pr, it might indeed be wrong offset

<stikonas[m]>I did see some evidence of it being wrong in gdb after that pr

<stikonas[m]>Anyway, I'll look more tomorrow

<oriansj>I trust you'll find the cleanest (least hacky way)

<stikonas[m]>Hmm, it would be a bit cleaner if depth was more consistent between arches, some arches have depth positive and some negative and then do add or sub compared to base pointer

<stikonas[m]>For no particular reason... Unless I'm mistakes, they all grow stack downwards

<stikonas[m]>s/mistakes/mistaken/

<stikonas>oriansj: just to make sure since I'm not super familiar with knigt. On knight the stack still grows down?

<stikonas>oriansj: https://github.com/oriansj/M2-Planet/pull/44

<stikonas>both global and local structs seem to work now

<stikonas>hmm, I've converted that PR to draft for now

<stikonas>somehow M2-Planet tests pass but stage0-posix doesn't seem to work

<stikonas>(except for x86)

<stikonas>(no, x86 also fails)

<stikonas>argh, that's unrelated, that's due to my dirty M2libc folder

<stikonas>indeed stage0-posix seems to work with my M2-Planet changes

IRC channel logs

2022-11-06.log