18:11 <lekernel> mwalle, do you have a contact to get my urjtag patch merged?

22:10 <Fallenou> hum I have some troubles with lm32 GNU as when trying to compile NetBSD kernel

22:13 <Fallenou> I get this error message from GNU as (called by GCC): /var/folders/ab/abyj3M5qFtefd2hEj876WU+++TI/-Tmp-//ccjcEt9q.s: Error: .size expression for compat_offseterr does not evaluate to a constant

22:13 <Fallenou> here is more information about the issue: http://pastebin.com/qy3gPC7s

22:13 <Fallenou> it seems my GNU as is having troubles with the ".size" gas operation

22:14 <Fallenou> all the warnings are "debug" I added inside GNU as to try to understand what goes wrong

22:15 <Fallenou> the assembly generated by gcc seems correct : .size symbol_name, .-symbol_name

22:15 <Fallenou> right after the end of the function

22:15 <Fallenou> so it seems to be a GNU AS only issue

22:20 <wpwrak> "Warning: This is the end!" such optimism ;-)

22:21 <Fallenou> the resolve_expression() of GAS seems to return 0 over there , leading to the Error : https://github.com/fallen/lm32-binutils-mmu/blob/master/gas/expr.c#L2197

22:22 <Fallenou> wpwrak: actually those messages mean that some of resolve_expression() call make it to the end of the function , to the return 1 meaning "all is ok" :)

22:22 <Fallenou> after verification left != right

22:23 <Fallenou> resolve_expression() is called by https://github.com/fallen/lm32-binutils-mmu/blob/master/gas/config/obj-elf.c#L1894

22:23 <Fallenou> called by elf_frob_symbol (what a strange name, I wonder what it means ...)

22:24 <Fallenou> and you can see a few lines below my error message "does not evaluate to a constant"

22:24 <wpwrak> maybe there's something weird in the function ?

22:24 <Fallenou> oh wait ...

22:25 <Fallenou> size->X_op is not O_constant

22:25 <Fallenou> it's O_subtract

22:25 <wpwrak> maybe try a code snippet from a (working) non-BSD build. then see if the .size works with the gas-for-BSD

22:26 <Fallenou> I wonder if resolve_expression() is called first ... (before the test of size->X_op == O_constant)

22:26 <wpwrak> if it does, take the problem code, then reduce it until it passes

22:26 <wpwrak> (reduce) first in C, then in assembler

22:28 <Fallenou> yeah you're right

22:28 <wpwrak> actually ... are those things even defined ?

22:28 <Fallenou> if ( f(x) && x->something == a ) <=== is f() called before the test x->something == a ?

22:29 <Fallenou> I would say yes

22:29 <Hawk777> Yes, because if f(x) returns false then the second expression is unevalutaed.

22:29 <Hawk777> Short circuit && operator.

22:29 <Fallenou> ok, that's what I would have said

22:30 <Fallenou> so size->X_op is not reduced by resolve_expression() to a constant, it stays a subtraction

22:30 <wpwrak> in the .s you pastebinned, i see no definition of compat_offseterr

22:30 <Fallenou> oh

22:30 <Fallenou> indeed

22:30 <Fallenou> I have mistaken the .global for a definition

22:30 <Fallenou> but it's not ...

22:31 <wpwrak> same for emul_find_root, etc.

22:31 <Fallenou> ok so it's GCC's fault :)

22:31 <Fallenou> gcc forgot to add "compat_offseterr: "

22:31 <wpwrak> so it seems :)

22:31 <Fallenou> great ! thanks :)

22:33 <wpwrak> or maybe it's one of the build options. -ffreestanding or -fno-common (i never used them)

22:40 <Fallenou> chaos

22:40 <Fallenou> ASM_DECLARE_FUNCTION_NAME ...

22:44 <Fallenou> ok it misses some "ASM_OUTPUT_LABEL"

22:48 <Fallenou> ok thanks for the help I was starting to lose my mind ^^

22:48 <Fallenou> gn8 !

22:53 <wpwrak> sane dreams ! ;-)