It's possibly an interaction between memoization and dynamic vars; more specifically, a result might be being memoized with one with-precision context active, and then recalled with a different one active, with arguments that compare equal despite the different precisions (say, because those arguments that are BigDecimals are exactly representable at the lower of the two precisions). If the result is *not* the same at the higher precision, that could cause an expected-equal check to fail in your test.
I'd also wonder at possibly having a test work if it used a memoized result generated earlier with a finite precision in effect, and fail with ArithmeticException without the memoization if the test doesn't set a precision limit itself and the result's not exactly representable at any finite precision (e.g. 1/7); but this would likely result in a test that worked in the suite and failed in isolation, instead of the reverse. However, if high memory use is causing a memoized value computed at limited precision to be "forgotten" and then recalculated with no precision limit, that might trigger failures in the suite of tests that worked in isolation. In any event, subtle interaction of with-precision's underlying dynamic var *math-context* with memoization is an obvious suspect worth investigating, even if it might ultimately be acquitted by the jury. :) On Mon, Jan 6, 2014 at 1:21 PM, Justin Kramer <jkkra...@gmail.com> wrote: > Shot in the dark: check that arguments passed to your memoized functions > use consistent typing. A BigDecimal such as 1M does not necessarily equal 1 > (a Long): > > > (= 1 1M) > false > > (== 1 1M) > true > > Your memoized functions could be recomputing values unnecessarily if > you're giving them different types of numbers. > > Justin > > > On Monday, January 6, 2014 12:57:49 PM UTC-5, David James wrote: >> >> I've got a Clojure test suite that fails when run as whole but passes >> when run piecemeal. This just started happening. I've tested the code >> thoroughly. The bug pops up in a part of the code that I did not change. >> So, at present, it feels like a heisenbug! >> >> These may be some relevant pieces: >> * I'm using memoization; the failures happen in memoized functions. >> * I'm doing a lot of various precision BigDecimal math. >> >> Some tentative guesses: >> * Somehow one test is affecting another test >> >> Questions (there are speculative): >> * What are some possible ways that memoization might fail (e.g. return an >> incorrect value)? >> * Is it possible that higher memory usage could cause memoization results >> to get lost? >> >> Unfortunately, it is a sizable code base, and I don't have a small, >> reproducible example. Any suggestions? >> >> -David >> > -- > -- > You received this message because you are subscribed to the Google > Groups "Clojure" group. > To post to this group, send email to clojure@googlegroups.com > Note that posts from new members are moderated - please be patient with > your first post. > To unsubscribe from this group, send email to > clojure+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/clojure?hl=en > --- > You received this message because you are subscribed to the Google Groups > "Clojure" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to clojure+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/groups/opt_out. > -- -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en --- You received this message because you are subscribed to the Google Groups "Clojure" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojure+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
