quickjs test suite fails on 32-bit systems due to size mismatches within `time_t` or related internal structs.

[Ashvith10]

While building for i686-linux locally on x86_64-linux:

$ ./pre-inst-env guix build --no-grafts quickjs-ng -s i686-linux
...
starting phase `check'
../build/run-test262 -c tests.conf
Result: 0/33 errors, 4 excluded
tests/test_builtin.js:690: Error: assertion failed: got |29256|, expected |29312| (order of operations / precision in MakeTime)

[saghul]

If you are building for i686 how are they being ran on x86_64?

[Ashvith10]

If you are building for i686 how are they being ran on x86_64?

@saghul Guix has a documentation on this section - Native Builds - taking x86_64 as the host architecture and i686 as the target architecture in the example.

The test results for a select few system are available here.

There's also a comment left out at:

quickjs/tests/test_builtin.js

Lines 687 to 690 in 97e6ef9

	// TODO: Fix rounding errors on Windows/Cygwin.
	if (!['win32', 'cygwin'].includes(os.platform)) {
	// from test262/test/built-ins/Date/UTC/fp-evaluation-order.js
	assert(Date.UTC(1970, 0, 1, 80063993375, 29, 1, -288230376151711740), 29312,

Platform is "linux" in this case and that is why it continues into the assertion and then fails.

[Ashvith10]

@saghul I've tried the flag-mpc64, but as it turns out, it fails later for the tests in run-test262 -m -c test262.conf -c test262-fast.conf -a

../build/run-test262 -m -c test262.conf -c test262-fast.conf -a
Result: 93/78324 errors, 3772 excluded, 5387 skipped, 10 new
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Atomics/pause/returns-undefined.js:22: unexpected error: TypeError: not an integral number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Atomics/pause/returns-undefined.js:22: strict mode: unexpected error: TypeError: not an integral number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum.js:11: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum.js:11: strict mode: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum-is-minus-zero.js:11: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum-is-minus-zero.js:11: strict mode: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum-is-NaN.js:10: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum-is-NaN.js:10: strict mode: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum-is-infinite.js:10: unexpected error: TypeError: not a number
/gnu/store/xqspqb4dk8fwfz7dv82lpihyc7c2w1jq-test262/test/built-ins/Math/sumPrecise/sum-is-infinite.js:10: strict mode: unexpected error: TypeError: not a number

Average memory statistics for 78324 tests:

QuickJS-ng memory usage -- 0.9.0 version, 32-bit, Little Endian, malloc limit: 0

NAME                    COUNT     SIZE
memory allocated         1128    82803  (73.4 per block)
memory used              1140    65743  (8 overhead, 15.0 average slack)
atoms                     603    23714  (39.3 per atom)
objects                   199     7973  (40.1 per object)
  properties             1011     8919  (5.1 per object)
  shapes                   71    14195  (199.9 per shape)
bytecode functions         14     6378
  bytecode                 14     1079  (77.1 per function)
  pc2line                  13      240  (18.5 per function)
C functions               118
arrays                      1
  fast arrays               1
  elements                  2       16  (2.0 per fast array)

make: *** [Makefile:129: test262-fast] Error 1

[Ashvith10]

After a little digging, I came across this. Maybe time will have to be handled depending on the system, as written in this SO page?

[bnoordhuis]

Can you check if the patch below fixes the sumPrecise errors for you?

diff --git a/quickjs.c b/quickjs.c
index 287194d..a4a7884 100644
--- a/quickjs.c
+++ b/quickjs.c
@@ -43239,7 +43239,7 @@ static JSValue js_math_sumPrecise(JSContext *ctx, JSValueConst this_val,
             goto fail;
         if (done)
             break; // item == JS_UNDEFINED
-        switch (JS_VALUE_GET_TAG(item)) {
+        switch (JS_VALUE_GET_NORM_TAG(item)) {
         default:
             JS_FreeValue(ctx, item);
             JS_ThrowTypeError(ctx, "not a number");

[saghul]

Ping?

[Ashvith10]

@saghul @bnoordhuis kindly ignore the previous comment - I've removed it as it was misleading. I've tried the patch, but the flags were incorrect, and it kept building for x86_64.

All I know from playing around the repo is that something related to JSValue's NaN boxing may be causing this issue.

Since this is beyond my expertise, unfortunately, I'll not be pursuing this issue anymore - but I'll leave it here in the open for anyone else who may be able to resolve this.

[suve]

I'm currently trying to package quickjs-ng for Fedora Linux and have also observed a test failure when building v0.11.0 for 32-bit x86.

+ ./redhat-linux-build/run-test262 -m -c test262.conf -a
Result: 94/79935 errors, 3285 excluded, 5569 skipped, 2 new
test262/test/built-ins/Atomics/pause/returns-undefined.js:22: unexpected error: TypeError: not an integral number
test262/test/built-ins/Atomics/pause/returns-undefined.js:22: strict mode: unexpected error: TypeError: not an integral number

[bnoordhuis]

Minor oversight, fix in #1235.