F9_sublattest fails.
Imported issue: Initially reported by @thome in https://gforge.inria.fr/tracker/?group_id=2065&aid=21374
I don't know whether I broke it. It seems it's been broken from the start.
CADO_DEBUG=1 CHECKS_EXPENSIVE=1 DEBUG=1 make -j8 F9_sublattest_dependencies && CADO_DEBUG=1 CHECKS_EXPENSIVE=1 DEBUG=1 make check ARGS=""-R F9_sublattest""
(well, you probably want to do all the F9 tests anyway because of that good ol' stupid misfeature).
Relation -113654,66679 is no longer printed.
trace_k output attached, but it does not make much sense.
I compared 6a80149a (old, birth of the sublat test) and 924769 (head of the sublat branch).
Note: two important differences.
-
the log scale has changed, because I dropped the ""maxlog2+=2"", which makes no sense to me.
-
the new lognorm code is more accurate.
So it could well be that rounding is at fault here, and it could play the other way around someday.
Also note that the sieve checksums have changed, although this per se is more probably related to the lognorm code having changed (even for the old commit, the check fails).
E.