[Regression] SIGSEGV on 1.6.2 on C function call, works on 1.0.0 up to 1.6.0

[arkanoid87]

Forum post:

https://forum.nim-lang.org/t/8796

Repo with minimal test case and C version to compare

https://github.com/arkanoid87/futhest

Re-posting forum thread here:

I've been scratching my head solving a SIGSEGV on my machine when using futhark. I've reduced the problem to a self contained Nim and C files with no external nim packages involved, just plain C linkage.

The C program compiles and runs correctly, the Nim program goes SIGSEGV while calling C function, but they should be 1:1

You can find dockerfiles for reproducible builds and Makefile here: https://github.com/arkanoid87/futhest.

shared header.h

int addition(int num1, int num2) {
    return num1+num2;
}

futhest.c

#include <assert.h>
#include <stdio.h>
#include "clang-c/Index.h"


enum CXChildVisitResult
visitor(CXCursor cursor, CXCursor parent, CXClientData clientData) {
    CXFile file;
    unsigned int line;
    unsigned int column;
    unsigned int offset;

    CXSourceLocation loc = clang_getCursorLocation(cursor);
    clang_getFileLocation(loc, &file, &line, &column, &offset);

    CXString filename = clang_getFileName(file);   
    printf("%s [%d:%d, %d]\n", clang_getCString(filename), line, column, offset);
    clang_disposeString(filename);

    return CXChildVisit_Continue;
}

int main (void) { 
    CXIndex Idx = clang_createIndex(0, 0);

    CXTranslationUnit TU = clang_parseTranslationUnit(Idx, 
        "src/header.h", NULL, 0, NULL, 0, 0);

    assert(TU != NULL);

    CXCursor cursor = clang_getTranslationUnitCursor(TU);

    clang_visitChildren(cursor, visitor, NULL);

    clang_disposeTranslationUnit(TU);

    printf("END");
}

C Output:

./futhest_c 
src/header.h [1:5, 4]
END⏎

futhest.nim

type
  CXIndex = pointer
  CXTranslationUnit = pointer
  CXCursor* {.bycopy.} = object
    kind*: byte
    xdata*: cint
    data*: array[3, pointer]

#[
CINDEX_LINKAGE CXIndex clang_createIndex(int excludeDeclarationsFromPCH,
                                         int displayDiagnostics);
]#
proc createIndex*(excludeDeclarationsFromPCH: cint; displayDiagnostics: cint): CXIndex {.
    importc: "clang_createIndex", cdecl.}


#[
CINDEX_LINKAGE CXTranslationUnit
clang_parseTranslationUnit(CXIndex CIdx,
                           const char *source_filename,
                           const char *const *command_line_args,
                           int num_command_line_args,
                           struct CXUnsavedFile *unsaved_files,
                           unsigned num_unsaved_files,
                           unsigned options);
]#
proc parseTranslationUnit*(CIdx: CXIndex; source_filename: cstring;
                          command_line_args: cstringArray;
                          num_command_line_args: cint;
                          unsaved_files: pointer;
                          num_unsaved_files: cuint; options: cuint): CXTranslationUnit {.
    importc: "clang_parseTranslationUnit", cdecl.}


#[
  CINDEX_LINKAGE CXCursor clang_getTranslationUnitCursor(CXTranslationUnit);
]#
proc getTranslationUnitCursor*(a1: CXTranslationUnit): CXCursor {.
    importc: "clang_getTranslationUnitCursor", cdecl.}


#[
  CINDEX_LINKAGE void clang_disposeTranslationUnit(CXTranslationUnit);
]#
proc disposeTranslationUnit*(a1: CXTranslationUnit) {.importc: "clang_disposeTranslationUnit", cdecl.}


# ----------------------------------------------------------------------------


var
  index = createIndex(0, 0)
  unit = parseTranslationUnit(index, "src/header.h".cstring, nil, 0, nil, 0, 0)

doAssert not unit.isNil

var cursor = getTranslationUnitCursor(unit) # SIGSEGV

# never reached
disposeTranslationUnit(unit)

echo "END"

Nim output:

./futhest
Traceback (most recent call last)
/home/jack/nim/futhest/src/futhest.nim(57) futhest
SIGSEGV: Illegal storage access. (Attempt to read from nil?)
fish: “./futhest” terminated by signal SIGSEGV (Address boundary error)

Additional Information

• 1.7.1 devel 9888a29: SIGSEGV
• 1.6.2: SIGSEGV
• 1.6.0: OK
• 1.4.8: OK
• 1.2.0: OK
• 1.0.0: OK

[ghost]

Isn't this a duplicate of #19342 ?

[ringabout]

I think so, too.

type
  CXCursor* {.bycopy.} = object
    kind*: byte
    xdata*: cint
    data*: array[3, pointer]

is passed by reference but shouldn't.

[arkanoid87]

I think so, too.

type
  CXCursor* {.bycopy.} = object
    kind*: byte
    xdata*: cint
    data*: array[3, pointer]

is passed by reference but shouldn't.

The SIGSEGV happens in cland_getTranslationUnitCursor(unit) function, specifically here https://github.com/llvm/llvm-project/blob/ef32c611aa214dea855364efd7ba451ec5ec3f74/clang/tools/libclang/CIndex.cpp#L4208
According to gdb

See PMunch/futhark#11 for some backlog

[ringabout]

It may be related: #19385

Sorry, I cannot test this issue locally, I don't get clang installed (I use windows and it is a hassle to use/install clang).

[arkanoid87]

It may be related: #19385

Sorry, I cannot test this issue locally, I don't get clang installed (I use windows and it is a hassle to use/install clang).

In the repository you find two Dockerfiles, one for Ubuntu and one for Alpine-linux. Both should work on Docker for windows, too (untested though)

Both replicate the problem

[ringabout]

In the repository you find two Dockerfiles, one for Ubuntu and one for Alpine-linux. Both should work on Docker for windows, too (untested though)

I see, thanks

[ghost]

Closing as a duplicate of #19342, as I said in https://forum.nim-lang.org/t/8796 that's the actual cause, not sure why you opened a separate issue. The crash happens in the libclang function - yes, but that's only because Nim passes an invalid number of arguments to it because it applies NRVO to a bycopy object.

From the generated C code you can see that the clang_getTranslationUnitCursor function is called with two arguments but it should in fact be called with one:

N_LIB_PRIVATE N_NIMCALL(void, NimMainModule)(void) {
{
	index__futhest_22 = clang_createIndex(((int) 0), ((int) 0));
	unit__futhest_23 = clang_parseTranslationUnit(index__futhest_22, "src/header.h", ((NCSTRING*) NIM_NIL), ((int) 0), NIM_NIL, ((unsigned int) 0), ((unsigned int) 0));
	{
		if (!!(!((unit__futhest_23 == 0)))) goto LA3_;
		failedAssertImpl__systemZassertions_56(((NimStringDesc*) &TM__FOvY5sgE9cCQaUMA1Y2m1uQ_2));
	}
	LA3_: ;
	clang_getTranslationUnitCursor(unit__futhest_23, (&cursor__futhest_29));
	clang_disposeTranslationUnit(unit__futhest_23);
	echoBinSafe(TM__FOvY5sgE9cCQaUMA1Y2m1uQ_3, 1);
}

[arkanoid87]

I've been impatient, you're right.

Thanks for the explanation.