Invalid memory access using the interpreter in file_spec

[beta-ziliani]

When executing crystal i spec/std/file_spec.cr (from the root of the project), it crashes after running successfully all the tests.

Maybe this is related to the fluke we see in CI?

My setup

Crystal 1.9.0-dev [1c186dc0e] (2023-06-29)

The compiler was not built in release mode.

LLVM: 15.0.7
Default target: x86_64-apple-macosx

Entire stacktrace

Invalid memory access (signal 11) at address 0x2d
[0x10d997e4b] *Exception::CallStack::print_backtrace:Nil +107 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10d960340] ~procProc(Int32, Pointer(LibC::SiginfoT), Pointer(Void), Nil)@src/crystal/system/unix/signal.cr:131 +304 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x7fff2089fd7d] _sigtramp +29 in /usr/lib/system/libsystem_platform.dylib
[0x10e89612a] *Crystal::Repl::Interpreter#interpret<Crystal::ASTNode+, Crystal::Type+>:Crystal::Repl::Value +375482 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10e839d61] *Crystal::Repl::Interpreter#interpret<Crystal::ASTNode+, Hash(String, Crystal::MetaVar)>:Crystal::Repl::Value +2289 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10eb8c2f3] *Crystal::Repl#interpret<Crystal::Call>:Crystal::Repl::Value +115 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10eb8c259] *Crystal::Repl#interpret_exit:Crystal::Repl::Value +41 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10eb8c208] *Crystal::Repl#run_file<String, Array(String)>:Crystal::Repl::Value +216 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10ed0bbb0] *Crystal::Command#repl:(Crystal::Repl::Value | Nil) +1440 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10ed05cc7] *Crystal::Command#run:(Bool | Crystal::Repl::Value | Nil) +1159 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10ed056cc] *Crystal::Command::run<Array(String)>:(Bool | Crystal::Repl::Value | Nil) +44 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10ed0564e] *Crystal::Command::run:(Bool | Crystal::Repl::Value | Nil) +46 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10d9362de] __crystal_main +2750 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10db44734] *Crystal::main_user_code<Int32, Pointer(Pointer(UInt8))>:Nil +36 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10db4468b] *Crystal::main<Int32, Pointer(Pointer(UInt8))>:Int32 +59 in /Users/beta/projects/crystal/crystal/.build/crystal
[0x10d9436b9] main +9 in /Users/beta/projects/crystal/crystal/.build/crystal

[straight-shoota]

I can reproduce this with compiler version 1c186dc. But this is an old version before 1.9.0.
With a current compiler build (1.9.2) there is no error. So this was probably fixed somewhere down the line.

[HertzDevil]

For a long time I thought this was a Windows-only issue, but I guess it isn't?

It looks like when the interpreter compiles the ::exit call that it invokes explicitly, the STDOUT inside the body is somehow different from the top-level one, so STDOUT gets initialized twice. Commenting out STDOUT.flush makes the spec no longer crash. I have no idea why STDERR doesn't have the same issue.