Regression on big-endian architectures

[kapouer]

quickjs-ng 0.16.2 fails its test suite on big-endian (s390x, ppc64, hppa, sparc64) with

cd build-static && ./run-test262 -vv -c ../tests.conf
../tests/bug488-upstream.js (1 ms)
../tests/bug39/2.js (2 ms)
../tests/bug633/0.js (0 ms)
../tests/bug39/1.js (1 ms)
../tests/assert.js (3 ms)
../tests/bug493.js (2 ms)
../tests/atomics-store-leak.js (5 ms)
../tests/bug39/3.js (1 ms)
../tests/bug633/1.js (1 ms)
../tests/bug567.js (3 ms)
../tests/bug633/3.js (0 ms)
run-test262: ./quickjs.c:2704: JS_FreeRuntime: Assertion `list_empty(&rt->gc_obj_list)' failed.
../tests/bug633/2.js (0 ms)
Aborted

I can have my hands on a s390x to get more info about that crash.

[kapouer]

NB: regression since 0.15.1.
https://buildd.debian.org/status/logs.php?pkg=quickjs-ng&arch=s390x

[kapouer]

More info when running it on s390x:

cd build-static && ./run-test262 -vv -c ../tests.conf
SyntaxError: checksum error
../tests/array-fromasync-enumerable.js:9: SyntaxError: checksum error
Object leaks:
       ADDRESS REFS SHRF          PROTO      CLASS PROPS
 0x3ffb402e350    1   0*  0x3ffb4015f90   Function { length: 0, name: 3"Array" }
 0x3ffb402e3a0    1   0*  0x3ffb4015f90   Function { length: 1, name: 4"TypeError" }
 0x3ffb402e3f0    1   0*  0x3ffb4015f90   Function { length: 3, name: "Object.defineProperty" }
run-test262: ./quickjs.c:2704: JS_FreeRuntime: Assertion `list_empty(&rt->gc_obj_list)' failed.
Aborted

[saghul]

Did you rebuild the bytecode?

[kapouer]

Because I could do that ? How ?
I just invoke cmake and make test. Build logs:
https://buildd.debian.org/status/fetch.php?pkg=quickjs-ng&arch=s390x&ver=0.16.2-1&stamp=1787241156&raw=0

[saghul]

For big endian you need you do make codegen first, since the bytecode is endian native, and the one we ship is little-endian.

[saghul]

I created #1681 for making this clear in the docs.

[bnoordhuis]

Answered, I believe? Closing.

[kapouer]

Yes, almost all green now: https://buildd.debian.org/status/package.php?p=quickjs-ng
(failure on hppa is not new).

[bnoordhuis]

I'll see if I can reproduce that crash locally with qemu-hppa.

[kapouer]

I'll see if I can reproduce that crash locally with qemu-hppa.

thanks, I opened an issue about that: #1698