I don't understand what the big pain point is, if Steve doesn't like the limit it he can just pick a different value no?
We do need some way of keeping the memory usage in check, people naturally want to show off largeish computations and we run the risk to require a beefy developer machine to be able to develop for Sage. As long as we have 32-bit buildbots the limit can't be much higher, but the writing is on the wall than 32-bit is going away eventually. >From my point of view its better to have some limit that provides immediate feedback during development than me having to maintain a special buildbot to limit memory usage and police tickets where developers don't have an easy way to reproduce. On Tuesday, March 23, 2021 at 1:21:43 AM UTC+1 Michael Orlitzky wrote: > Is anyone very much in love with the --memlimit (default: 3300MB) > option to the sage -t command? > > Once again it has completely broken testing on some systems. We could > try to guess a new, higher limit... or just admit that maybe it's not a > great idea after all and delete the thing. The latter is what I'm about > to propose on https://trac.sagemath.org/ticket/31395 > > A few points: > > * The failure of the default limit has been and will always be a > recurring problem as sage requires more and more memory. Every > time we hit it, a bunch of machines are broken until the limit > can be raised in the "develop" branch. > > * The default memory limit exists for precisely one doctest (which > I've refactored to still be tested, modulo the next bullet point). > > * Testing out-of-memory conditions doesn't test what you'd expect, > since if you _actually_ run out of memory, all hell breaks loose > on the system at the same time as your graceful error handling > kicks in. > > * A global limit is likely incorrect, as would be revealed if there > were more than one doctest using it. Each test needs the limit to > be low enough to trigger a failure, but not low enough to crash > the rest of sage. Both of these numbers are test- and system- > dependent. > > * Reimplementing "ulimit -v" in a mathematics suite is a waste of > development resources. > > I'd rather just delete it and generalize the one existing doctest with > something like a "with memlimit(...)" context manager in the unlikely > event that we ever have another test for OOM behavior. > > > -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/536abe7e-c9fd-4336-aa03-8d430508a472n%40googlegroups.com.
