Exported int sometimes corrupted with -sMAIN_MODULE

[hoodmane]

Sometimes references a C variable via HEAP32[_variable/4] doesn't work.

a.c

const int my_number = 123456;

b.c

extern const int my_number;
#include "stdio.h"

int main(void) {
    printf("my_number: %d\n", my_number);
}

pre.js

Module.preRun = () => {
    console.log("HEAP32[_my_number/4]:", HEAP32[_my_number/4]);
}

Compile, link, execute

emcc -fPIC -c a.c
emcc -fPIC -c b.c

LDFLAGS="\
    -sEXPORTED_FUNCTIONS=_main,_my_number \
    -sMAIN_MODULE=1 \
    --pre-js=pre.js \
"

emcc $LDFLAGS b.o a.o
node a.out.js

Edit: To reproduce, b.o comes before a.o.

Output

I got:

HEAP32[_my_number/4]: 1701209717
my_number: 123456

It should print:

HEAP32[_my_number/4]: 123456
my_number: 123456

Changes that fix it

1. Drop -sMAIN_MODULE or -sMAIN_MODULE=2.
2. Swap the order of the object files a.o and b.o so that the link command is emcc $LDFLAGS a.o b.o

If I drop the const in a.c, then it seems to return the same wrong number independent of the link order.

Version of emscripten/emsdk

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 3.1.72-git (65f3d78fed2d1786afe278ec20bbe13425d8a51c)
clang version 20.0.0git (https:/github.com/llvm/llvm-project 50866e84d1da8462aeb96607bf6d9e5bbd5869c5)
Target: wasm32-unknown-emscripten
Thread model: posix

and

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 3.1.68 (ceee49d2ecdab36a3feb85a684f8e5a453dde910)
clang version 20.0.0git (https:/github.com/llvm/llvm-project 5cc64bf60bc04b9315de3c679eb753de4d554a8a)
Target: wasm32-unknown-emscripten
Thread model: posix

[sbc100]

I can't seem to reproduce this on tot:

$ cat build.sh 
./emcc -fPIC -c a.c
./emcc -fPIC -c b.c

LDFLAGS="\
    -sEXPORTED_FUNCTIONS=_main,_my_number \
    -sMAIN_MODULE=1 \
    --pre-js=pre.js \
"

./emcc $LDFLAGS a.o b.o
node a.out.js
$ sh build.sh 
emcc: warning: EXPORTED_FUNCTIONS is not valid with LINKABLE set (normally due to SIDE_MODULE=1/MAIN_MODULE=1) since all functions are exported this mode.  To export only a subset use SIDE_MODULE=2/MAIN_MODULE=2 [-Wunused-command-line-argument]
HEAP32[_my_number/4]: 123456
my_number: 123456

[hoodmane]

I am still getting this on tip of tree:

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 3.1.72-git (079e2660071d5ae7d45d8e04efc09dd7e67456dd)

[hoodmane]

I'm sorry, I got the order to link the object files wrong in the original report.
Here's a self contained script:

#!/bin/bash
rm -rf build && mkdir build && cd build
echo "const int my_number = 123456;" > a.c
cat << EOF > b.c
extern const int my_number;
#include "stdio.h"

int main(void) {
    printf("my_number: %d\n", my_number);
}
EOF
cat << EOF > pre.js
Module.preRun = () => {
    console.log("HEAP32[_my_number/4]:", HEAP32[_my_number/4]);
}
EOF

emcc -fPIC -c a.c
emcc -fPIC -c b.c


LDFLAGS="\
    -sEXPORTED_FUNCTIONS=_main,_my_number \
    -sMAIN_MODULE=1 \
    --pre-js=pre.js \
"

if test $# -eq 0; then
    echo "This works fine:"
    emcc $LDFLAGS a.o b.o
else
    echo "This does not work:"
    emcc $LDFLAGS b.o a.o 
fi
node a.out.js

If you run it with no arguments, you get the correct output. If you run it with one or more arguments, you get the wrong output.

[hoodmane]

@sbc100 could you try again to see if you can reproduce this?

[sbc100]

Yes, I have been able to reproduce. I simplified a little removing the pre.js file:

$ cat test.sh 
#!/bin/bash
echo "const int my_number = 123456;" > num.c
cat << EOF > main.c
#include "stdio.h"
#include <emscripten/em_asm.h>
extern const int my_number;

int main(void) {
  EM_ASM(console.log("JS:_my_number:", _my_number, HEAP32[_my_number/4]));
  printf("C: my_number: %ld %d\n", (long)&my_number, my_number);
}
EOF

set -e
set -x

./emcc -fPIC -c main.c
./emcc -fPIC -c num.c

LDFLAGS="-sEXPORTED_FUNCTIONS=_main,_my_number -sMAIN_MODULE=1"

./emcc $LDFLAGS main.o num.o
node a.out.js
$ sh test.sh 
+ ./emcc -fPIC -c main.c
+ ./emcc -fPIC -c num.c
+ LDFLAGS='-sEXPORTED_FUNCTIONS=_main,_my_number -sMAIN_MODULE=1'
+ ./emcc -sEXPORTED_FUNCTIONS=_main,_my_number -sMAIN_MODULE=1 main.o num.o
emcc: warning: EXPORTED_FUNCTIONS is not valid with LINKABLE set (normally due to SIDE_MODULE=1/MAIN_MODULE=1) since all functions are exported this mode.  To export only a subset use SIDE_MODULE=2/MAIN_MODULE=2 [-Wunused-command-line-argument]
+ node a.out.js
JS:_my_number: 3660 1634878572
C: my_number: 22128 123456

[sbc100]

An even simpler repro with just a single source file:

#!/bin/bash
cat << EOF > main.c
#include "stdio.h"
#include <emscripten.h>
#include <emscripten/em_asm.h>

EMSCRIPTEN_KEEPALIVE int my_number = 123456;

int main(void) {
  EM_ASM(console.log("JS:_my_number:", _my_number, HEAP32[_my_number/4]));
  printf("C: my_number: %ld %d\n", (long)&my_number, my_number);
}
EOF

set -e
set -x

./emcc -sMAIN_MODULE main.c
node a.out.js

I found that issue and have a fix.