Bug: Failed to raise an exception: END_OF_STACK

[asterite]

Code to reproduce:

class Foo
  def initialize
    @buffer = Pointer(UInt8).malloc(17)
    header = @buffer.as({Int32, Int32, Int32}*)
    header.value = {String::TYPE_ID, 4, 0}
  end

  def to_s
    @buffer.as(String)
  end
end

string = Foo.new.to_s
data = string.to_slice
if data[0, 4].to_a != [] of UInt8
  raise "OH NO!"
end

You have to compile it with --release, then when you run it you get:

Failed to raise an exception: END_OF_STACK
[0x1040f3f27] *CallStack::print_backtrace:Int32 +39
[0x1040ccc90] __crystal_raise +48
[0x1040d2468] *raise<NilAssertionError>:NoReturn +136
[0x1040d2361] *Nil#not_nil!:NoReturn +17
[0x1040f3b97] __crystal_sigfault_handler +55
[0x7fff79632b5d] _sigtramp +29

This was extracted from a larger code that uses String.build, which essentially does the same as Foo.

I don't know what's going on, maybe stack corruption of some sort.

[ysbaddaden]

It seems to break when trying to print the backtrace. What happens if you disable it?

[kostya]

similar error on linux and mac, with Crystal 0.31.0 (2019-09-24) LLVM: 8.0.1 Default target: x86_64-apple-macosx with pipes.

crystal --version | echo

Error writing file: Broken pipe (Errno)
  from IO::FileDescriptor@Crystal::System::FileDescriptor#unbuffered_write<Slice(UInt8)>:Nil
  from IO::FileDescriptor@IO::Buffered#flush:IO::FileDescriptor
  from exit<Int32>:NoReturn
  from exit:NoReturn
  from Crystal::Command#run:(Bool | Crystal::Compiler::Result | Nil)
  from __crystal_main
  from main
Error: you've found a bug in the Crystal compiler. Please open an issue, including source code that will allow us to reproduce the bug: https://github.com/crystal-lang/crystal/issues
Unhandled exception: Error writing file: Broken pipe (Errno)
  from IO::FileDescriptor@Crystal::System::FileDescriptor#unbuffered_write<Slice(UInt8)>:Nil
  from IO::FileDescriptor@IO::Buffered#flush:IO::FileDescriptor
  from exit<Int32>:NoReturn
  from Crystal::error<String, Bool, Int32, IO::FileDescriptor, Bool>:Nil
  from Crystal::error:exit_code<String, Bool, Int32>:Nil
  from Crystal::Command#error<String, Int32>:Nil
  from Crystal::Command#error<String>:Nil
  from Crystal::Command#run:(Bool | Crystal::Compiler::Result | Nil)
  from __crystal_main
  from main
Failed to raise an exception: END_OF_STACK
[0x100b03b17] *CallStack::print_backtrace:Int32 +39
[0x100129b70] __crystal_raise +48
[0x100146a71] *IO::FileDescriptor@Crystal::System::FileDescriptor#unbuffered_write<Slice(UInt8)>:Nil +1057
[0x10012dc4f] main +143

[Blacksmoke16]

Someone reported this for oq as well Blacksmoke16/oq#39.

[asterite]

Might be related to #4276

[bcardiff]

This might get fixed by #8728

[jwoertink]

I just compiled that code example on 0.33 and 0.34 in both standard and release build and they worked fine; however, this doesn't.

[12:56PM] sandbox$ crystal --version | echo

Error writing file: Broken pipe (IO::Error)
  from IO::FileDescriptor@Crystal::System::FileDescriptor#unbuffered_write<Slice(UInt8)>:Nil
  from IO::FileDescriptor@IO::Buffered#write<Slice(UInt8)>:Nil
  from Crystal::Command#run:(Bool | Crystal::Compiler::Result | Nil)
  from __crystal_main
  from main
Error: you've found a bug in the Crystal compiler. Please open an issue, including source code that will allow us to reproduce the bug: https://github.com/crystal-lang/crystal/issues
[12:56PM] sandbox$ crystal --version
Crystal 0.34.0 (2020-04-06)

LLVM: 6.0.1
Default target: x86_64-apple-macosx
[12:57PM] sandbox$ ruby --version
ruby 2.4.2p198 (2017-09-14 revision 59899) [x86_64-darwin19]
[12:58PM] sandbox$ ruby --version | echo

So maybe the original issue is fixed, but there's another underlying issue?

[RX14]

This isn't a bug that broken pipe errors exist. I'm glad they display properly now though.

[jwoertink]

@RX14 The error I posted isn't a bug? If that's the case, then should I open an issue to say that getting that error shouldn't tell us to open an issue about it? On a side note, I can't upgrade to Crystal 0.34 because my Lucky app keeps crashing. This error shows up a lot in my logs, so my thought was that it was related. I don't really understand the error though, so I have no idea.

[waj]

@jwoertink I think it's totally a bug. It's not bug in the compiler, though. Here's a minimal example that also fails:

# foo.cr:
puts "Hello"

$ crystal foo.cr | echo

Unhandled exception: Error writing file: Broken pipe (IO::Error)
  from /usr/local/Cellar/crystal/0.34.0/src/crystal/system/unix/file_descriptor.cr:82:13 in 'unbuffered_write'
  from /usr/local/Cellar/crystal/0.34.0/src/io/buffered.cr:218:5 in 'flush'
  from /usr/local/Cellar/crystal/0.34.0/src/io/buffered.cr:179:7 in 'write_byte'
...

Could you please fill a separate issue for this?

[waj]

I'm sorry, I keep forgetting this is not a bug, @RX14 is right and I even wrote this not so long ago: #7810 (comment)

@jwoertink is right though, that the compiler shouldn't crash that way because the stdout is closed. And of course, the message is clear 🙂

I'd like to see those errors with Lucky to understand if that's related in any way.

[jwoertink]

@waj https://forum.crystal-lang.org/t/unhandled-exception-error-writing-file-broken-pipe-errno/1396/4 These are the main top of the errors. Beyond those lines just gets in to app specific path stuff.

I'm curious though, I can run that line of code through ruby, and even node without any errors. Do they just swallow that exception? I haven't tried another compiled language, so maybe that's the difference?

[waj]

@jwoertink let's talk in the issue you just opened so we keep the topic organized for future reference