test/run-pass/core-rt-smoke crashes on Windows

[crnobog]

Starting program: c:\code\rust\i686-pc-mingw32\test\run-pass\core-rt-smoke.stage2-i686-pc-mingw32.exe
[New Thread 8284.0x2488]

Program received signal SIGSEGV, Segmentation fault.

0x680d04d3 in rt::args::load_argc_and_argv::_2199da865b9a2acc::_0$x2e8$x2dpre () from c:\code\rust\install\bin\std-6c65cf4b443341b1-0.8-pre.dll

(gdb) disassemble
Dump of assembler code for function

_ZN2rt4args18load_argc_and_argv17_2199da865b9a2acc14_0$x2e8$x2dpreE:
   0x680d0470 <+0>:     cmp    %fs:0x14,%esp
   0x680d0477 <+7>:     ja     0x680d0489 <_ZN2rt4args18load_argc_and_argv17_2199da865b9a2acc14_0$x2e8$x2dpreE+25>
   0x680d0479 <+9>:     push   $0xc
   0x680d047e <+14>:    push   $0x24
   0x680d0483 <+19>:    call   0x680d42c8 <__morestack>
   0x680d0488 <+24>:    ret
   0x680d0489 <+25>:    push   %ebp
   0x680d048a <+26>:    mov    %esp,%ebp
   0x680d048c <+28>:    push   %ebx
   0x680d048d <+29>:    push   %edi
   0x680d048e <+30>:    push   %esi
   0x680d048f <+31>:    sub    $0x14,%esp
   0x680d0492 <+34>:    movl   $0x18,0x4(%esp)
   0x680d049a <+42>:    call   0x6803acd0 <_ZN4libc5funcs3c956stdlib6malloc16_54e7a493f6e1cfe14_0$x2e8$x2dpreE>
   0x680d049f <+47>:    mov    %eax,%esi
   0x680d04a1 <+49>:    test   %esi,%esi
   0x680d04a3 <+51>:    jne    0x680d04aa <_ZN2rt4args18load_argc_and_argv17_2199da865b9a2acc14_0$x2e8$x2dpreE+58>
   0x680d04a5 <+53>:    call   0x680dc328 <abort>
   0x680d04aa <+58>:    movl   $0x0,(%esi)
   0x680d04b0 <+64>:    movl   $0x10,0x4(%esi)
   0x680d04b7 <+71>:    mov    %esi,-0x10(%ebp)
   0x680d04ba <+74>:    cmpl   $0x0,0xc(%ebp)
   0x680d04be <+78>:    je     0x680d055f <_ZN2rt4args18load_argc_and_argv17_2199da865b9a2acc14_0$x2e8$x2dpreE+239>
   0x680d04c4 <+84>:    xor    %ebx,%ebx
   0x680d04c6 <+86>:    lea    -0x10(%ebp),%esi
   0x680d04c9 <+89>:    nop
   0x680d04ca <+90>:    nop
   0x680d04cb <+91>:    nop
   0x680d04cc <+92>:    nop
   0x680d04cd <+93>:    nop
   0x680d04ce <+94>:    nop
   0x680d04cf <+95>:    nop
   0x680d04d0 <+96>:    mov    0x10(%ebp),%eax
=> 0x680d04d3 <+99>:    mov    (%eax,%ebx,4),%eax
   0x680d04d6 <+102>:   cmpb   $0x0,(%eax)
   0x680d04d9 <+105>:   mov    $0x0,%ecx
   0x680d04de <+110>:   je     0x680d04fa <_ZN2rt4args18load_argc_and_argv17_2199da865b9a2acc14_0$x2e8$x2dpreE+138>
   0x680d04e0 <+112>:   xor    %ecx,%ecx

The code before the call above:

Dump of assembler code for function _ZN2rt5start15_98ebfd32a7b8f114_0$x2e8$x2dpreE:
=> 0x68082e80 <+0>:     cmp    %fs:0x14,%esp
   0x68082e87 <+7>:     ja     0x68082e99 <_ZN2rt5start15_98ebfd32a7b8f114_0$x2e8$x2dpreE+25>
   0x68082e89 <+9>:     push   $0x14
   0x68082e8e <+14>:    push   $0x2c
   0x68082e93 <+19>:    call   0x680d42c8 <__morestack>
   0x68082e98 <+24>:    ret
   0x68082e99 <+25>:    push   %ebp
   0x68082e9a <+26>:    mov    %esp,%ebp
   0x68082e9c <+28>:    push   %edi
   0x68082e9d <+29>:    push   %esi
   0x68082e9e <+30>:    and    $0xfffffff8,%esp
   0x68082ea4 <+36>:    sub    $0x20,%esp
   0x68082ea7 <+39>:    mov    0x18(%ebp),%edi
   0x68082eaa <+42>:    mov    0x14(%ebp),%esi
   0x68082ead <+45>:    mov    0x10(%ebp),%eax
   0x68082eb0 <+48>:    mov    %eax,0x8(%esp)
   0x68082eb4 <+52>:    mov    0xc(%ebp),%eax
   0x68082eb7 <+55>:    mov    %eax,0x4(%esp)
   0x68082ebb <+59>:    call   0x680d0470 <_ZN2rt4args18load_argc_and_argv17_2199da865b9a2acc14_0$x2e8$x2dpreE>
   0x68082ec0 <+64>:    mov    %eax,0x4(%esp)

(gdb) x/8xw $ebp

0x28fe40:       0x0028fe70      0x68082ec0      0x00480000      0x00400000
0x28fe50:       0x00000000      0x30e6b812      0xfffffffe      0x770e6b96

The instruction mov 0x10(%ebp),%eax looks like it's trying to load argv which is a null. But the value before that doesn't look like a sensible value of argc, so the failure must have happened earlier in the initialisation of the process or the runtime.

[brson]

This code should never be called on windows or mac. It's support code for std::os that is only supposed to be used on linux. Both args::init and cleanup should be conditionally-compiled.

[crnobog]

test/run-fail/tls-exit-status crashes in the same place when using RUST_NEWRT=1, I assume that'll be fixed by this as well.

[alexcrichton]

@crnobog, out of curiosity, can you confirm whether this test is now passoing or failing after the std::rt::args module is conditionally compiled out for windows?

[alexcrichton]

This test no longer exists, closing.