Re: regression on zOS between 5.4.8 to 5.5.0 on zOS

Hi all.

Eric Covener you can check/show lua stack (elements, size) between 
versions  ?





Em domingo, 8 de fevereiro de 2026 às 22:02:43 UTC-3, gottfried leibniz 
escreveu:

> On 2/6/2026 5:13 PM, Eric Covener wrote:
> > We are on the old/traditional native xlc. Our HTTPD (+prereqs) port is
> > somewhat old (~2005), so we are not using any of the more modern
> > porting-friendly compilers or options.
>
> Interesting.
>
> I spent time testing 5.5.0 on older z/OS images w/ 'V1.10 XL C/C++',
> 'V2.4 XL C/C++', and 'C/C++ for Open Enterprise Languages on z/OS 2.0.0,
> clang version 14.0.0' and could not replicate this issue (note: all
> older z/OS compilers).
>
> Given that your Lua compiler listing output resembles what happens when
> using older nvidia C/C++ compilers[1], I would experiment with making
> the temp variable in dischargevars[2] volatile to see if things improve.
> Although, given that prior releases of Lua work for you, I would be
> surprised if it did.
>
> In nvc's case, everything in testes/ passes after applying that
> universal band-aid/hack.
>
> [1] For reference, compiling lcode.c (w/ HEAD at c6b484823) using nvc
> 22.11-0 will emit for [2]:
>
> 199f: ff 24 c5 00 00 00 00 jmp *0x0(,%rax,8) // VLOCAL 
> to 19AD
> ....:
> 19ad: 41 c7 06 08 00 00 00 movl $0x8,(%r14) // VNONRELOC
> 19b4: e9 d9 02 00 00 jmp 1c92 
> <luaK_dischargevars+0x312>
> ....:
> 1c92: 48 83 c4 08 add $0x8,%rsp // Function 
> epilogue
>
> Meanwhile, with nvc 26.1-0:
>
> 1a1f: ff 24 c5 00 00 00 00 jmp *0x0(,%rax,8) // VLOCAL 
> to 1A2D
> ....:
> 1a2d: 0f b6 43 08 movzbl 0x8(%rbx),%eax
> 1a31: 89 43 08 mov %eax,0x8(%rbx)
> 1a34: c7 03 08 00 00 00 movl $0x8,(%rbx) // VNONRELOC
> 1a3a: e9 ce 02 00 00 jmp 1d0d 
> <luaK_dischargevars+0x30d>
> ....:
> 1d0d: 48 83 c4 08 add $0x8,%rsp // Function 
> epilogue
>
> AFIAK this was fixed in nvhpc-23-1.
>
> [2] 
>
> https://github.com/lua/lua/blob/c6b484823806e08e1756b1a6066a3ace6f080fae/lcode.c#L829
>

-- 
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/1c1b0da7-a0d4-40e8-b628-8576f7f63d02n%40googlegroups.com.