Just a guess, recently we switched from gcc to clang. clang is too smart in optimization that it short circuit some code that deals with NaN, I've put some inline asm to teach clang register flags need checking. This works in the past.
Can you try build and test with -O0 or with gcc? On Tue, Aug 7, 2018, 9:48 AM Henry Rich <[email protected]> wrote: > This baffles me. I wonder if there is some compiler interaction with > _clearfp(). In case so, I have pushed out another change that puts a > couple of memory fetches between the clearfp()s. Would you please build > that & report back? If it fails on the same line, please try the two > parts individually & see if they both fail. > > Henry Rich > > > On 8/6/2018 6:13 PM, Bill Heagy wrote: > > 32-bit again? > > > > ...... > > 1 > > 'domain error' -: +/@:*"1~ etx u: 'abc' > > 1 > > > > NB. Test for NaN > > 0 _ 1 (+/@:*"1 -: +/@:*"_"1) _ 0 2 NB. removable NaN > > |NaN error > > | 0 _ 1 (+/@:*"1-:+/@:*"_"1)_ 0 2 > > |[-189] /home/wheagy/git/jsource/test/g420fg.ijs > > > > > > JVERSION > > Engine: j806/j32/linux > > Beta: GPL3/2018-08-06T17:48:48 > > Library: 8.05.14 > > Platform: Linux 32 > > Installer: unknown > > InstallPath: /home/wheagy/tmp/jbld/j32 > > Contact: [email protected] > > ---------------------------------------------------------------------- > > For information about J forums see http://www.jsoftware.com/forums.htm > > > --- > This email has been checked for viruses by AVG. > https://www.avg.com > > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
