search

JohnCremona / eclib

The eclib package includes mwrank (for 2-descent on elliptic curves over Q) and modular symbol code used to create the elliptic curve database.
GNU General Public License v2.0
21 stars 15 forks source link

Version 20220621 started failing on S390X recently #73

Closed SnarkBoojum closed 2 years ago

SnarkBoojum commented 2 years ago

The same version which was working is now failing. I'm not sure exactly why, but since pari got updated in Debian recently, it's a possible explanation.

Since it's a timeout thing, perhaps something takes a little more time and that's enough to break... how can I help debug the issue?

JohnCremona commented 2 years ago

Looking at the log, the only test which starts but does not finish is tsatbnd (looking for "Testing abc..." with no "abc completed", taking into account that they can be interleaved). It is true that the saturation code (which this tests) was changed quite a lot in 2021 (in fact as a result of saturation now being also implemented in libpari -- though that is not used here).

During those changes I added one more call to libpari; before this it only used libpari for integer factorization, while now it also uses the ellap() and ellinit() functions, and these are called during the saturation process.

Is it possible to see the partial output file tsatbnd.testout whcih would be created in the tests/ directory (but normally deleted at the end of the tests)? It seems likely that the sticking point is in the new large test I added adfter making the saturation process much more efficient...

SnarkBoojum commented 2 years ago

I found no issue on the porter box ; so I asked for a new build of the package, which went through flawlessly: that was a machine hiccup, not a real problem...

Sorry for the noise!