Compiler segfaults after printing error in some tests

[marijnh]

I observed this in src/test/compile-fail/extft-non-literal.rs and src/test/compile-fail/extft-non-literal2.rs , might be happening in others. This is probably an unwinding bug.

With some changes to our generated code (I was working on making calls simpler), it hangs instead of segfaulting when extft-non-literal2.rs fails.

[brson]

I don't see this, and a segfaulting compiler should cause tests to fail now. Was the crash in map perchance? I've seen that a number of times.

[marijnh]

What happens when you try to compile one of these from the command-line? For me, it prints out the error, then segfaults with this stack trace:

#0  0x082d097f in glue_drop208 ()
#1  0x081d5dc1 in syntax::fold::noop_fold_local ()
#2  0x0830fdeb in syntax::fold::make_fold::thunk5295 ()
#3  0x0830cf72 in syntax::fold::noop_fold_decl::anon5136 ()
#4  0xf7f0e129 in vec::map () from stage1/libstd.so
#5  0x081d0821 in syntax::fold::noop_fold_decl ()
#6  0x0830f555 in syntax::fold::make_fold::thunk5285 ()
#7  0x081cf5e3 in syntax::fold::noop_fold_stmt ()
#8  0x0830f235 in syntax::fold::make_fold::thunk5282 ()
#9  0xf7f0e129 in vec::map () from stage1/libstd.so
#10 0x081cf311 in syntax::fold::noop_fold_block ()
#11 0x0830f13f in syntax::fold::make_fold::thunk5281 ()
#12 0x081d4efe in syntax::fold::noop_fold_fn ()
#13 0x0830fa74 in syntax::fold::make_fold::thunk5289 ()
#14 0x081cdd9d in syntax::fold::noop_fold_item_underscore ()
#15 0x0830efae in syntax::fold::make_fold::thunk5279 ()
#16 0x081cd81e in syntax::fold::noop_fold_item ()
#17 0x0830ef57 in syntax::fold::make_fold::thunk5278 ()
#18 0xf7f0e129 in vec::map () from stage1/libstd.so
#19 0x081d53fa in syntax::fold::noop_fold_mod ()
#20 0x0830fadd in syntax::fold::make_fold::thunk5290 ()
#21 0x081cc087 in syntax::fold::noop_fold_crate ()
#22 0x0830ebcf in syntax::fold::make_fold::thunk5274 ()
#23 0x0823b5a8 in syntax::ext::expand::expand_crate ()
#24 0x0833c769 in driver::rustc::compile_input::thunk7959 ()
#25 0x082a139d in driver::rustc::time ()
#26 0x082a2cd5 in driver::rustc::compile_input ()
#27 0x082bc567 in driver::rustc::main ()
#28 0x082c61b4 in _rust_main ()
#29 0x0834358c in _rust_main_wrap ()
#30 0xf7fa9b31 in task_start_wrapper (a=0xf6b9e010) at ./src/rt/rust_task.cpp:149
#31 0xdeadbeef in ?? ()

[brson]

You are right, and the test runner isn't looking for a specific non-zero return value when it runs the compiler. I wonder how many of these compile failures are actually segfaults...

[brson]

The anwser is 18. 18 compile-fail tests are segfaulting.

[brson]

Also opened #938 to make the test runner catch this in the future

[marijnh]

So this is not a symptom of #939 then?

[brson]

#939 did fix one of them

[marijnh]

Fair enough. I'm going to xfail a few that hang instead of segfaulting with a patch I'm trying to land. Will refer to this bug in the patch, so that we can be sure to re-enable the tests when the problem is found.

[brson]

Now 14 failures:

    [compile-fail] ../src/test/compile-fail/ext-nonexistent.rs
    [compile-fail] ../src/test/compile-fail/extenv-no-args.rs
    [compile-fail] ../src/test/compile-fail/extenv-not-string-literal.rs
    [compile-fail] ../src/test/compile-fail/extenv-too-many-args.rs
    [compile-fail] ../src/test/compile-fail/extfmt-missing-type.rs
    [compile-fail] ../src/test/compile-fail/extfmt-no-args.rs
    [compile-fail] ../src/test/compile-fail/extfmt-non-literal.rs
    [compile-fail] ../src/test/compile-fail/extfmt-not-enough-args.rs
    [compile-fail] ../src/test/compile-fail/extfmt-unknown-type.rs
    [compile-fail] ../src/test/compile-fail/extfmt-unsigned-plus.rs
    [compile-fail] ../src/test/compile-fail/extfmt-unsigned-space.rs
    [compile-fail] ../src/test/compile-fail/extfmt-unterminated-conv.rs
    [compile-fail] ../src/test/compile-fail/macro-2.rs
    [compile-fail] ../src/test/compile-fail/macro.rs