Segmentation fault when using string created from empty array

[aturley]

System

OS:

OSX 10.11.6

compiler:

0.10.0-7c7e036 [release]
compiled with: llvm 3.8.1 -- Apple LLVM version 7.3.0 (clang-703.0.29)

Issue

Observed Behavior

Using a string that is created using String.from_array where the array is empty results in a segmentation fault.

Expected Behavior

A string that is created using String.from_array should behave like a zero-length string.

Steps to Reproduce

This program reproduces the issue:

actor Main
  new create(env: Env) =>
    let s = String.from_array(recover Array[U8] end)
    env.out.print("before")
    env.out.print("'" + s + "'")
    env.out.print("after")

Additional information

I ran the reproduction program (above) in lldb. Here's the backtrace after it segfaults:

(lldb) bt
* thread #2: tid = 0x859eae, 0x00007fff8aafafd6 libsystem_platform.dylib`_platform_memmove$VARIANT$Haswell + 182, stop reason = EXC_BAD_ACCESS (code=1, address=0x0)
  * frame #0: 0x00007fff8aafafd6 libsystem_platform.dylib`_platform_memmove$VARIANT$Haswell + 182
    frame #1: 0x0000000100003855 string-from-empty-array`Pointer_U8_val_box__copy_to_oZo + 21
    frame #2: 0x000000010000366c string-from-empty-array`String_iso__append_oo(this=0x00000001097d6480, s=0x00000001097d6460) + 124 at string.pony:1108
    frame #3: 0x000000010000335d string-from-empty-array`String_val_add_oo(this=0x0000000100014fc0, that=0x00000001097d6460) + 141 at string.pony:1118
    frame #4: 0x0000000100002f7b string-from-empty-array`Main_tag_create_oo(this=0x00000001097ff000, env=0x00000001097fe800) + 283 at main.pony:5
    frame #5: 0x00000001000015a5 string-from-empty-array`Main_Dispatch + 117
    frame #6: 0x0000000100004d92 string-from-empty-array`handle_message + 530
    frame #7: 0x000000010000e084 string-from-empty-array`run_thread + 1332
    frame #8: 0x00007fff8b28499d libsystem_pthread.dylib`_pthread_body + 131
    frame #9: 0x00007fff8b28491a libsystem_pthread.dylib`_pthread_start + 168
    frame #10: 0x00007fff8b282351 libsystem_pthread.dylib`thread_start + 13
(lldb) 

[jemc]

Looks like String._append is still assuming that String always has a null byte at the end.

We need someone to rewrite the logic in String._append to be aware of the possibility that not all strings are internally represented as being null-terminated. And we'll also want tests that prove correct behaviour in these circumstances, both for the case of zero-length string being appended, and for the case of an immutable string slice with no null-terminator, that was created via String.trim.