IRC channel logs

<matrix_bridge><gtker> ekaitz: With "--dirty-mode" it won't delete intermediate files and you can inspect them. The M1 files are the closest to assembly

<ekaitz>good!

<ekaitz>thanks

<stikonas>--dirty mode I guess is for M2-Mesoplanet

<stikonas>well, M2-Planet indeed will just assemble, it won't do anythign more

<stikonas>and that assembly will have a few comments too, so will help you understand where you are

<ekaitz>yeah

<ekaitz>i just saw

<ekaitz>i have the same function compiled with the ifs and without them

<ekaitz>but ufff i don't know x86

<stikonas>well, instructions that we use are not that different from riscv

<stikonas>but you could use riscv build too for this :D

<ekaitz>yeah...

<ekaitz>what i did is to convert my switch back to an if but with the same structure

<ekaitz>and it fails

<ekaitz>so it's me that I didn't understand the code path properly

<ekaitz>stikonas: does this make sense for you? https://paste.debian.net/1363338/

<ekaitz>it doesn't make any sense to me

<ekaitz>i just repeated the code using switches instead of the if and the values are different

<ekaitz>and taking some example

<ekaitz>SWITCH___________________________________________

<ekaitz>1011666872 1014045680 475200264 536466608 538845416 1011666872 1014045680

<ekaitz>IFS______________________________________________

<ekaitz>1011666872 1014045680 475200264 536466608 538845416 536466608 538845416

<ekaitz>the cdr value, the last one

<ekaitz>1014045680 - 538845416 = 475200264

<ekaitz>so, it's using the d value without the - dist

<ekaitz>so it's falling in the default case when it shouldn't

<ekaitz>it's falling always in the default case

<ekaitz>:(

<ekaitz>stikonas: i'd say the switch is broken

<stikonas>ekaitz: perhaps...

<stikonas>it was just developed yesterday...

<ekaitz>the other switch works, because the default is empty

<ekaitz>well, no

<ekaitz>idk why the other one works, maybe it's because this one has numbers and the other has characters

<ekaitz>idk

<stikonas>strange though

<stikonas>default was also tested here https://github.com/oriansj/M2-Planet/blob/master/test/test0031/switch.c

<ekaitz>i bet it can be reproduced as is

<ekaitz>the t value is 0,1,2...

<ekaitz>what else can it be? the switch there is so obvious...

<ekaitz>oh wait

<ekaitz>hm

<ekaitz>the only case that doesn't fall in the default is the 0 case

<ekaitz>which should fall in the default

<stikonas>well, default is implemented here https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_core.c#L2821

<ekaitz>stikonas: i found the problem

<ekaitz>i replaced the constants with the actual value

<ekaitz>and now it worked

<ekaitz>so the defines are not being expanded

<ekaitz>also the enums don't work (i tried)

<stikonas>oh, I see

<stikonas>still better than nothing I guess

<stikonas>hmm, let me check, there is https://github.com/oriansj/M2-Planet/blob/master/test/test0031/switch.c#L91 in the test

<ekaitz>stikonas: got it

<ekaitz> case :TMACRO·

<ekaitz>·

<ekaitz> case :7

<ekaitz>that's how it expands

<stikonas>is TMACRO not 7?

<ekaitz>shouldn't the columns be after the number?

<ekaitz>no no, those are two cases

<ekaitz>the 7 i wrote by hand

<ekaitz>the TMACRO y just left

<ekaitz>the preprocessing should replace those shouldn't it?

<ekaitz>oh maybe not because I see some STACK_SIZE out there too

<ekaitz>in any case, why is the colon before the number?

<ekaitz>i just run -E

<stikonas>those colons are processed by M2-Planet tokenizer

<ekaitz>so the preprocessor converts C to something else?

<stikonas>well, just this thing

<stikonas>it's artifact from label treatment

<stikonas>C labels are "label:"

<stikonas>but M1 labels are ":label"

<stikonas>hence I put https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_core.c#L952 back as workaround

<ekaitz>but :7 is the same as :HELLO?

<ekaitz>because the first works and the second goes to default

<stikonas>where is this :HELLO?

<stikonas>in switch.c?

<ekaitz>it's just an example

<stikonas>or your code

<stikonas>oh yeah, I think so

<stikonas>unless we missed soemthing

<ekaitz>in my code the defines don't work but the number does

<stikonas>so that if statement only works on numbers

<stikonas>so maybe that's why

<stikonas>see https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_core.c#L954C3-L954C40

<stikonas>so what happens to enums then...

<ekaitz>enums don't work either

<stikonas>ok, so clearly bug in M2-Planet

<stikonas>I migth not have time to fix it until end of March though

<stikonas>maybe gtker would though

<ekaitz>with the enums el output is the same

<ekaitz>:NAME_OF_THE_ENUM

<stikonas>ekaitz: yeah, so I think what is happening

<stikonas>is for :LABEL

<stikonas>we are hitting https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_core.c#L950C10-L950C35

<stikonas>but typedefs and enums are processed in the next block

<stikonas> https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_core.c#L962

<ekaitz>hm and what does that do?

<stikonas>if it notices :7 it returns 7

<stikonas>but see next block deals with variables and then block after that with numbers

<stikonas>so I think that if : conditional

<stikonas>should actually just strip : and pass to the next 2 blocks (variables and numbers)

<ekaitz>hm!

<stikonas>so I think I mostly understand why it happens

<stikonas>we can release M2-Planet 1.12.1 later once this is fixed...

<ekaitz>:)

<ekaitz>i may try to fix it but I have short time blocks and it's going to be hard

<matrix_bridge><gtker> stikonas: Are you working on a patch or do you want me to open a PR?

<stikonas>well, I tried a bit of experimenting, but didn't get anything working

<stikonas>feel free to try...

<stikonas>I don't have much time today anyway

<stikonas>but I think the problem is in constant_unary_expression in the handling of :

<matrix_bridge><gtker> Is there a repro example that fails?

<stikonas>ekaitz: also note that those constant expressions (in case and in other places such as array size) don't respect operator priority

<stikonas>but this is probably an edge case

<stikonas>gtker probably even switch.c in our testsuite

<stikonas>e.g. https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/test/test0031/switch.c#L91 that you added

<matrix_bridge><gtker> Is there a failing program I can test against? As far as I can see we're using an enum in the switch.c tests

<stikonas>gtker: but it's switch(10)

<stikonas>so it never gets processed

<stikonas>but anyway, I don't have any testcase either

<stikonas>ekaitz: can you help gtker with testcase?

<matrix_bridge><gtker> stikonas: I've got one it seems.

<matrix_bridge><gtker> int test_default() {

<matrix_bridge>int a = 6;

<matrix_bridge>switch (a) {

<matrix_bridge> case SIX:

<matrix_bridge> return 0;

<matrix_bridge> default: return 1;

<matrix_bridge>}

<matrix_bridge>return 1;

<matrix_bridge>}

<matrix_bridge><gtker> Fails on current

<stikonas>that inverse : behavious is somewhat annoying :(

<stikonas>gtker: maybe it is easier to undo that?

<stikonas>and fix https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_core.c#L3365 ?

<ekaitz>oh sorry guys I was going to send you one but I was busy

<ekaitz>so yeah

<ekaitz>that's consistent with what I experienced

<matrix_bridge><gtker> stikonas: Yeah, it might be. At least for case values since they aren't labels. Might be annoying since we need to add state to the tokenizer for when the previous token was "case"

<stikonas>well, I was thinking about undoing it completely

<stikonas>tokenize : separately

<stikonas>and then for labels, check if next token is :

<stikonas>instead of first char is :

<matrix_bridge><gtker> Hmm, yeah, that would also work and might be cleaner. It would probably require relatively large changes though?

<stikonas>but where else : is used?

<stikonas>just labels and case

<matrix_bridge><gtker> Yeah, true

<stikonas>and part in tokenizer is https://github.com/oriansj/M2-Planet/blob/754004e1b7d89b0f311447f1515190fff51da357/cc_reader.c#L279

<matrix_bridge><gtker> I believe I have a slightly hacky workaround for now. But it's probably a good idea to fix labels

<matrix_bridge><gtker> I'll add a few more tests and create a PR. Probably within 20 minutes

<matrix_bridge><gtker> stikonas, ekaitz: PR is up https://github.com/oriansj/M2-Planet/pull/103