You might try running science-test.rkt multiple times and see if it fails. It's the simulation collection where I see the problem mostly. It uses the differential equation solver from the science collection, which is the only place I know of that uses boxes, but there may be others. I was using Windows 7 64-bit today when I tried it. I'll see if science-test works for me. When it does fail for me now, it is consistently with a seg fault in gc. I'll keep checking nightly builds on my home machine in the evenings. I'll look into the unsafe-fl* error. Thanks, Doug On Sunday, August 28, 2011, Matthew Flatt <<a href="mailto:mflatt@cs.utah.edu">mflatt@cs.utah.edu</a>> wrote: > At Sun, 28 Aug 2011 13:18:36 -0600, Doug Williams wrote: >> I reported a problem a few weeks back that keeps me from using 5.1.2 (and >> now 5.1.3). Matthew believes it is due to an optimizer bug with boxes. >> Fortunately, 5.1.1 works, but we're stuck there. I tried the latest nightly >> build this morning and the error is still there. > > I thought I had fixed that one with push d185c2a0df45 in the git repo. > > The line > > (require (planet "williams/science/examples/science-test.ss")) > > works for me (Mac OS X, 32-bit Windows, 64-bit Windows). Am I trying > the right example? > > >> Anyway, the real reason for this e-mail is that the latest nightly build >> (which was plt-3.1.3.6-bin-x86_64-win32.exe) gave me another problem. When I >> downloaded the science collection from PLaneT using the latest nightly >> build, it gave me errors about expecting a module for some files. There are >> indeed some rkt files that are included - using include - that are not >> modules. But, they are in the compile-omit-files in the info.rkt file. Is >> this being deprecated in some way? > > I see that problem (i.e., errors when I use the `require' line above), > and it's related to recent changes for `raco link'. I've pushed a > repair in time for tonight's build. > > > BTW, the bytecode optimizer notices that `unsafe-fl*' is applied to > three arguments in "statistics.rkt". Looks like it's on line 713. > >