Re: regression on zOS between 5.4.8 to 5.5.0 on zOS

On 2/4/2026 7:10 AM, Eric Covener wrote:
> Big endian is the obvious one but it shares that with AIX which is unaffected.
> I have also seen warnings that "strict aliasing" is rather
> strict/aggressive on zOS / xlc.
> 

What z/OS compiler is being used here? In your debug output, one of the
OP_GETTABUP instructions [1] looks different when compiling the included
example on other machines (RA_Raw *should* be 4).

If you are building Lua from the release page, would it be possible to
generate and provide the compiler listing, e.g., ./luac -l -l -p $FILE
(It may also be helpful to include both 5.5 and 5.4.8).

> I also maintain this stack on AIX, another big endian platform, and
 > see no problem with 5.5.0.

While I presently do not have have access to a z/OS box (or one modern
enough with the requisite LLVM PTFs for Open XL C/C++), two related
issues I have hit in the past:

1. Older XL C/C++ 2.4.1 compilers (xlc/xlclang) will incorrectly
optimize the table.unpack loop[2] even with the strict options set
(i.e., -qstrict + -qstrict_induction), leading to a segfault around one
of the functions edge cases[3].

2. Older Nvidia C/C++ compilers (e.g., nvc 22.11-0) struggled with
overlapping assignments in lcode.c (even with [4] applied). For example,
compiling your case with this buggy compiler will output:

         5       [18]    SELF            0 0 255
         6       [18]    LOADK           2 5     ; "Hello Lua World: "
         7       [18]    GETTABUP        3 0 6   ; _ENV "type"

Which is also wrong and has an identical RA_Raw for OP_GETTABUP to your
case.

[1] 
https://gist.github.com/covener/4412c8be7762916335bb05e1d8171511#file-lua-txt-L36
[2] 
https://github.com/lua/lua/blob/2a7cf4f319fc276f4554a8f6364e6b1ba4eb2ded/ltablib.c#L214-L216
[3] 
https://github.com/lua/lua/blob/2a7cf4f319fc276f4554a8f6364e6b1ba4eb2ded/testes/sort.lua#L98
[4] 
https://github.com/lua/lua/commit/d51022bf9e496ae4a7276b600d2755becc7d4323

-- 
You received this message because you are subscribed to the Google Groups "lua-l" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
To view this discussion visit https://groups.google.com/d/msgid/lua-l/adc80935-cea7-4e5b-a1c1-1aab47e8ed18%40gmail.com.