cpp file "undefined reference to"

[mobilemindtec]

Hello,

I'm trying scala-native and doing some testing. How do I add .cpp to my project?

If I create a lib.c file with the content:

int add3(int i) {
    return i + 3;
}

Everything works normally. But if I take the same file and rename it to lib.cpp I get the following compilation error:

[error] /usr/bin/ld: /mnt/data/sources/scala/native/app/target/scala-3.3.1/native/6.ll.o: in function `_SM5Main$D4mainLAL16java.lang.String_uEO':
[error] 6.ll:(.text+0xf797): undefined reference to `add3'
[error] /usr/bin/ld: 6.ll:(.text+0xf7a3): undefined reference to `cprintln'
[error] clang: error: linker command failed with exit code 1 (use -v to see invocation)
[error] Failed to link /mnt/data/sources/scala/native/app/target/scala-3.3.1/app-out
[error] (Compile / nativeLink) Failed to link /mnt/data/sources/scala/native/app/target/scala-3.3.1/app-out
[error] Total time: 20 s, completed Jan 18, 2024, 5:26:44 PM

I didn't find any instructions in the documentation about this. Am I missing something?

My scala code:

@extern
object lib {
  def add3(i: CInt) CInt = extern
}

I'm already try add .withClangPP(file("/usr/bin/clang++").toPath), but not work.

[ekrich]

You need to wrap your function with extern "C". See https://stackoverflow.com/questions/1041866/what-is-the-effect-of-extern-c-in-c

[mobilemindtec]

OMG this work! Thank you very much!

[ekrich]

Great! I think this would be a good item for the docs.

[mobilemindtec]

Hi,

I'm advancing a little further in my Scala-Native studies. So I decided to make a simplified wrap version of boost-beast. When I used a single file httpserver.cpp using extern "C" for the run function, it worked. But when I added a little more complexity, with some .cpp and other .h files, the compilation stopped working, with several errors about std:

error: no member named 'optional' in namespace 'std'

I already tried adding withCompileOptions(c.compileOptions ++ Seq("-std=c++17")), but compile return this error:

[error] error: invalid argument '-std=c++17' not allowed with 'C'

If it helps, here is the project source: https://github.com/mobilemindtec/scala-beast

What am I missing?

[ekrich]

I think since our compileOptions go to both C and C++ the C one is complaining about C++ option. Our current compliance is c++11 that is added automatically to only C++ files being compiled. So right now this is not supported.

[mobilemindtec]

Got it, thanks for the explanation. Therefore, I was able to adjust just by adjusting/reimplementing the parts incompatible with cpp+11. Now app worked. I'll continue testing and see if this turns into something! Thanks again!

Here is the result of the scala on the beast.

Statistics        Avg      Stdev        Max
  Reqs/sec      5004.51    1947.44   10903.24
  Latency      592.69ms   154.52ms      3.05s
  Latency Distribution
     50%   548.01ms
     75%   574.16ms
     90%   660.42ms
     95%      0.85s
     99%      1.36s
  HTTP codes:
    1xx - 0, 2xx - 152979, 3xx - 0, 4xx - 0, 5xx - 0
    others - 39

[mobilemindtec]

I have a strange situation. When I configure the server to run multithread, the application freezes. Something related to the callback provided by scala. My test callback function is "pure", it doesn't access any shared resources, It just creates the response, which is a struct. No error occurs, the server just stops responding. When there is no scala callback, the server does not freeze. Will this be a limitation of the current version? Any idea? Thanks!

[WojciechMazur]

When I configure the server to run multithread, the application freezes.

No idea what is happening inside the server, maybe attaching with debuger (lldb, gdb) might give us a hints where it's blocked.
If underlying boost threads would create concurrent threads accessing Scala code then might lead to data races in GC and possibly create illegal state. SN 0.4.x does not support multithreading and is not prepared for concurrent allocations in GC (these might not be visible in user code at first glance, becouse we might need to introduce argument boxing in some places). For multithreading support you can try to use 0.5.0-SNAPSHOT, altough it currently does not expose an API allowing to attach non-managed threads (threads not created via java.lang.Thread)
If so you can test you app with custom GC, e.g. conservative Boehm GC

[mobilemindtec]

This is beyond my knowledge, but it really seems to be an issue related to concurrent memory allocation:

Thread 1 "http-native-out" received signal SIGSEGV, Segmentation fault.
__memset_avx2_unaligned_erms () at ../sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:244
Downloading source file /usr/src/debug/glibc/glibc/string/../sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
244             VMOVU   %VMM(0), (%rdi)                                                                                     
(gdb) where
#0  __memset_avx2_unaligned_erms () at ../sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:244
#1  0x00005555557b80d1 in _SM21java.lang.StackTrace$D28currentStackTrace$$anonfun$1L36scala.collection.mutable.ArrayBufferL29scala.scalanative.unsafe.ZoneuEPT21java.lang.StackTrace$ ()
#2  0x000055555576bfb5 in _SM31java.lang.StackTrace$$$Lambda$2D9applyVoidL16java.lang.ObjectuEO ()
#3  0x00005555557410e8 in _SM34scala.runtime.function.JProcedure1D5applyL16java.lang.ObjectL16java.lang.ObjectEO ()
#4  0x000055555576a534 in _SM30scala.scalanative.unsafe.Zone$D5applyL15scala.Function1L16java.lang.ObjectEO ()
#5  0x00005555557b7beb in _SM21java.lang.StackTrace$D17currentStackTraceLAL27java.lang.StackTraceElement_EO ()
#6  0x00005555557b69f3 in _SM19java.lang.ThrowableD16fillInStackTraceL19java.lang.ThrowableEO ()
#7  0x000055555576ec1b in _SM34scala.scalanative.runtime.package$D14throwClassCastR_R_nEO ()
#8  0x000055555577e64a in _SM16httpserver.Main$D11handlerSyncL28scala.scalanative.unsafe.PtrL28scala.scalanative.unsafe.PtrEO ()
#9  0x000055555577e7c7 in _SM16httpserver.Main$D36runCSyncServer$$anonfun$1$$anonfun$1L28scala.scalanative.unsafe.PtrL28scala.scalanative.unsafe.PtrEPT16httpserver.Main$ ()
#10 0x000055555573b07c in _SM26httpserver.Main$$$Lambda$2G17$extern$forwarder ()
#11 0x000055555565cfbd in httpserver::http_handler::dispatch (this=0x5555558e43c0, req=0x5555558e4930)
    at /mnt/data/sources/scala/native/http-native/target/scala-3.3.1/native/native-code-classes-0/scala-native/http_handler.cpp:37
#12 0x00005555556b53be in httpserver::http_session::handle_request<boost::beast::http::basic_string_body<char, std::char_traits<char>, std::allocator<char> >, std::allocator<char> > (this=0x5555558e26d0, req=...)
    at /mnt/data/sources/scala/native/http-native/target/scala-3.3.1/native/native-code-classes-0/scala-native/httpserver.cpp:345
#13 0x00005555556885d5 in httpserver::http_session::on_read (this=0x5555558e26d0, ec=..., bytes_transferred=62)
    at /mnt/data/sources/scala/native/http-native/target/scala-3.3.1/native/native-code-classes-0/scala-native/httpserver.cpp:89

[WojciechMazur]

Looks like a data race, probably 1 thread allocated class X, other one allocated at overlapping location class. Same thing probably happend in recursive currentStackTrace. There's nothing we can do in 0.4.x. The only alternative would be to use Boehm GC, and somehow force the boost threads to run GC_register_my_thread at start and make sure that GC_THREADS is defined, see https://stackoverflow.com/questions/20968114/boehm-gc-with-c11s-thread-library

[mobilemindtec]

where can I find version 0.5.0-SNAPSHOT?