crash when logging tuple with bare strings

[tedhorst]

The following program compiles fine, but segfaults when run (as of a52f6d2):

fn main() {
    log(error, ("hi there!", "you"));
}

Backtrace is:

Thread 2 Crashed:: scheduler loop 0
0   librustrt.dylib                 0x00000001017643e0 shape::log::walk_string2(std::pair<shape::ptr, shape::ptr> const&) + 94 (rust_shape.cpp:384)
1   librustrt.dylib                 0x000000010176454e shape::log::walk_vec2(bool, std::pair<shape::ptr, shape::ptr> const&) + 48 (rust_shape.cpp:426)
2   librustrt.dylib                 0x000000010176789d shape::ctxt<shape::data<shape::log, shape::ptr> >::walk_slice0() + 143 (rust_shape.h:1198)
3   librustrt.dylib                 0x0000000101767ee1 shape::ctxt<shape::data<shape::log, shape::ptr> >::walk() + 1565 (rust_shape.h:318)
4   librustrt.dylib                 0x0000000101764b84 shape::log::walk_struct2(unsigned char const*) + 106 (rust_shape.cpp:416)
5   librustrt.dylib                 0x0000000101767d03 shape::ctxt<shape::data<shape::log, shape::ptr> >::walk() + 1087 (rust_shape.h:477)
6   librustrt.dylib                 0x0000000101764708 shape_log_type + 264 (rust_shape.cpp:530)
7   librustrt.dylib                 0x000000010176c936 __morestack + 9
8   librustrt.dylib                 0x000000010175c745 rust_task::call_on_c_stack(void*, void*) + 155 (rust_task.h:463)
9   librustrt.dylib                 0x000000010175d7df upcall_log_type + 159 (rust_upcall.cpp:35)
10  log_bare_str                    0x00000001011b7ca5 main::_908aa4b47b687ff::_00 + 133
11  log_bare_str                    0x00000001011b7cde _rust_main + 46
12  librustrt.dylib                 0x000000010175c355 task_start_wrapper(spawn_args*) + 37 (rust_task.cpp:155)

[tedhorst]

As of 4f323d0 and the change of logging to visitor, the above works, but now the following does not:

fn main() {
    log(error, (123_i32, ~"hi there!"));
}

Backtrace is:

* thread #2: tid = 0x2403, 0x0000000100076525 libcore-c3ca5d77d81b46c1-0.4.dylib`repr::__extensions__::meth_11435::visit_estr::_bb314cd7b2d5487c::_04 + 117, stop reason = EXC_BAD_ACCESS (code=13, address=0x0)
    frame #0: 0x0000000100076525 libcore-c3ca5d77d81b46c1-0.4.dylib`repr::__extensions__::meth_11435::visit_estr::_bb314cd7b2d5487c::_04 + 117
    frame #1: 0x000000010200540f
    frame #2: 0x000000010000ed6d libcore-c3ca5d77d81b46c1-0.4.dylib`repr::__extensions__::meth_4985::visit_estr_uniq::_bb314cd7b2d5487c::_04 + 109
    frame #3: 0x000000010000236b log_bare_str`glue_visit_1814 + 75
    frame #4: 0x000000010000ff48 libcore-c3ca5d77d81b46c1-0.4.dylib`repr::__extensions__::meth_5029::visit_tup_field::_4a8896f1ceaa562::_04 + 184
    frame #5: 0x0000000100002274 log_bare_str`glue_visit_1808 + 260
    frame #6: 0x00000001000038bb log_bare_str`intrinsic::rusti::visit_tydesc_1924::_b9735e93cd22bed::_00 + 75
    frame #7: 0x0000000100001cfe log_bare_str`repr::write_repr_1776::_1f9fb77bcf4d318::_00 + 702
    frame #8: 0x0000000100001a34 log_bare_str`logging::log_type_1773::anon + 68
    frame #9: 0x00000001000041df log_bare_str`__morestack + 224
    frame #10: 0x0000000100001a14 log_bare_str`logging::log_type_1773::anon + 36
    frame #11: 0x0000000100032c1d libcore-c3ca5d77d81b46c1-0.4.dylib`io::with_bytes_writer::_95ef8a6f639649c::_04 + 189
    frame #12: 0x00000001000018c7 log_bare_str`logging::log_type_1773::_46883f83cbac29e::_00 + 103
    frame #13: 0x0000000100001617 log_bare_str`main::_c7f570f6468c2ea::_00 + 215
    frame #14: 0x00000001000016ae log_bare_str`_rust_main + 46
    frame #15: 0x00000001001fb0f5 librustrt.dylib`task_start_wrapper(spawn_args*) + 37 at rust_task.cpp:150

[catamorphism]

Works now, thanks to the removal of shape code.

[tedhorst]

The second example still fails in visitor code.

[catamorphism]

Sorry, didn't test the second example! I'll take a look.

[graydon]

I'll take this. It's a bug in the reflection code. The string isn't the issue; the i32 is. Also fauls with i8, i16, but not i64. Padding calculation is wrong somewhere.

[graydon]

Oh. We're still hitting write_repr, not the post-transition target write_repr2. The former doesn't do alignment right at all. This is very fixable.

[graydon]

This is closed, github just doesn't know it yet.