Bug#1017962: mozjs102: Fails to build on armhf

[Jeremy Bicha]

Source: mozjs102
Version: 102.2.0-1
Severity: important
X-Debbugs-CC: debian-***@lists.debian.org
Tags: ftbfs help

mozjs102 fails to build on armhf. This is a problem because it is
blocking us from being able to switch Debian's gjs to use mozjs102
instead of the obsolete mozjs91.

https://buildd.debian.org/status/logs.php?pkg=mozjs102&arch=armhf

Here's an excerpt from the build log:

In function ‘uint64_t js::jit::AtomicCmpXchg64SeqCst(uint64_t*,
uint64_t, uint64_t)’,
inlined from ‘static T
js::jit::AtomicOperations::fetchAndSeqCst(T*, T) [with T = long long
int]’ at /<<PKGBUILDDIR>>/js/src/jit/shared/AtomicOperations-shared-jit.h:337:1,
inlined from ‘static T
js::jit::AtomicOperations::fetchAndSeqCst(SharedMem<T*>, T) [with T =
long long int]’ at
/<<PKGBUILDDIR>>/js/src/jit/AtomicOperations.h:212:26,
inlined from ‘virtual bool
cls_testAtomicOperationsI64::run(JS::HandleObject)’ at
/<<PKGBUILDDIR>>/js/src/jsapi-tests/testAtomicOperations.cpp:223:3:
/<<PKGBUILDDIR>>/debian/build/js/src/jit/AtomicOperationsGenerated.h:240:17:
error: ‘asm’ operand has impossible constraints
240 | asm volatile (
| ^~~


Thank you,
Jeremy Bicha

[Simon McVittie]

Post by Jeremy Bicha
mozjs102 fails to build on armhf

...

Post by Jeremy Bicha
In function ‘uint64_t js::jit::AtomicCmpXchg64SeqCst(uint64_t*,
uint64_t, uint64_t)’,
inlined from ‘static T
js::jit::AtomicOperations::fetchAndSeqCst(T*, T) [with T = long long
int]’ at /<<PKGBUILDDIR>>/js/src/jit/shared/AtomicOperations-shared-jit.h:337:1,
inlined from ‘static T
js::jit::AtomicOperations::fetchAndSeqCst(SharedMem<T*>, T) [with T =
long long int]’ at
/<<PKGBUILDDIR>>/js/src/jit/AtomicOperations.h:212:26,
inlined from ‘virtual bool
cls_testAtomicOperationsI64::run(JS::HandleObject)’ at
error: ‘asm’ operand has impossible constraints
240 | asm volatile (

I'm doing a build on amdahl to get the contents of the generated header,
but it's going to take a while to get there.

The firefox source package is basically the same codebase and it compiled
successfully on armhf, which might be a useful data point?

smcv

[Simon McVittie]

Post by Simon McVittie

Post by Jeremy Bicha
mozjs102 fails to build on armhf

...

Post by Jeremy Bicha
In function ‘uint64_t js::jit::AtomicCmpXchg64SeqCst(uint64_t*,
uint64_t, uint64_t)’,
inlined from ‘static T
js::jit::AtomicOperations::fetchAndSeqCst(T*, T) [with T = long long
int]’ at /<<PKGBUILDDIR>>/js/src/jit/shared/AtomicOperations-shared-jit.h:337:1,
inlined from ‘static T
js::jit::AtomicOperations::fetchAndSeqCst(SharedMem<T*>, T) [with T =
long long int]’ at
/<<PKGBUILDDIR>>/js/src/jit/AtomicOperations.h:212:26,
inlined from ‘virtual bool
cls_testAtomicOperationsI64::run(JS::HandleObject)’ at
error: ‘asm’ operand has impossible constraints
240 | asm volatile (

I'm doing a build on amdahl to get the contents of the generated header,
but it's going to take a while to get there.

Here's the generated file. The build is still going, so I don't know
whether the FTBFS will be reproducible.

smcv

[Simon McVittie]

Control: forwarded -1 https://bugzilla.mozilla.org/show_bug.cgi?id=1761665

Post by Simon McVittie
Here's the generated file. The build is still going, so I don't know
whether the FTBFS will be reproducible.

Looks like yes. I think this is the same issue as
https://bugzilla.mozilla.org/show_bug.cgi?id=1761665 upstream,
so I'm trying the workaround suggested there.

smcv