Nanojit's 'make check' currently uses the default --random size, which is a 1000 instruction block. This should be made much bigger, to really stress things -- most of the problems I've seen with --random have only manifested with bigger sizes. I suggest 100,000 which takes about 0.2s with a debug build on my fast Linux box. I'd also be happy with even higher, eg. 1,000,000, esp. since the NJ tests are extremely fast anyway.
Created attachment 420225 [details] [diff] [review] patch I upped it to 1,000,000. If we can let the machine do the work, we might as well really stress it. (Motivation for doing this now is that we just found a lirasm assert that wasn't being triggered with smaller random sizes.)
http://hg.mozilla.org/projects/nanojit-central/rev/759fd4ffc918 Note that this change doesn't need to be merged to TM and TR because it only affects the NJ-central build system. So I'm marking it as closed.
There's also the testlirc.sh script, which is used in the TM copy. I updated that as well: http://hg.mozilla.org/projects/nanojit-central/rev/3f09056fb0a7 Reopening bug.