[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-08-06 Thread Olivier Tilloy
Some progress: running one of the failing tests through nodejs with the --no-opt flag (for no optimizations) makes it consistently pass. It looks like V8 is optimizing too aggressively, and incorrectly. I can observe that when values start diverging, ft() is being called with s===2, and despite

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-08-06 Thread Olivier Tilloy
Some notes while I continue investigating the problem: - while I used the last test in test/test.js to highlight the problem, it isn't the only one that is failing - both sha.js and sha1.js are affected (unsurprisingly, as they have very similar implementations), other algorithms (sha224,

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-08-03 Thread Olivier Tilloy
Unfortunately no, the patch doesn't affect the outcome of tests on ppc64el (I did try reverting it, and the problem has also been reported upstream, which doesn't carry the patch). -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu.

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-07-31 Thread Bryce Harrington
Could it have anything to do with the patch being applied in node- sha.js, 0002-Fix-FTBFS-in-32-bits-arch.patch? var hash = new Sha1() - var bigData = Buffer.alloc(0x1 / 8) + var bigData = Buffer.alloc(0x1fff8 / 8) I wonder if this patch were disabled for ppc64el, if the test

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-07-31 Thread Olivier Tilloy
(the above comment applies to the last test in test/test.js with the label "call digest for more than MAX_UINT32 bits of data". -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1887144 Title:

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-07-31 Thread Olivier Tilloy
Added some logging to the Sha1._update() method which is called repeatedly, and the problem doesn't appear to be deterministic: it doesn't always start diverging at the same place (although it's always early, within the first 100 iterations). -- You received this bug notification because you are

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-07-30 Thread Olivier Tilloy
** Description changed: Since nodejs 12.18.1 entered groovy-proposed, autopkgtests that rely on - node-sha to compute SHA-1 hashes started failing reliably on ppc64el. + node-sha.js to compute SHA-1 hashes started failing reliably on ppc64el. This affects the following packages:   -

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-07-10 Thread Olivier Tilloy
** Package changed: node-sha (Ubuntu) => node-sha.js (Ubuntu) ** Description changed: Since nodejs 12.18.1 entered groovy-proposed, autopkgtests that rely on node-sha to compute SHA-1 hashes started failing reliably on ppc64el. This affects the following packages: - - node-sha - -

[Bug 1887144] Re: autopkgtest failures on ppc64el with nodejs 12.18.1

2020-07-10 Thread Olivier Tilloy
And testing in a ppc64el cloud instance, I reverted nodejs to 10.19.0 ~dfsg-3ubuntu1 and the tests reliably pass. This confirms that something in node-sha broke with the update to nodejs 12.18.1~dfsg-1ubuntu1. -- You received this bug notification because you are a member of Ubuntu Bugs, which