Kasper Daniel Hansen wrote: > On Jul 11, 2007, at 9:30 AM, Prof Brian Ripley wrote: <snipped> >>> The full Makevars file: >>> >>> MYCXXFLAGS=-O0 >> You do realize that is highly non-portable? > > Yes, but I do not know of any other way to downgrade the optimization > level. The SDK I am using breaks (due to memory alignment issues if a > specific optimization flag for GCC is include - the flag is certainly > implied by -O2 and as far as I recall also -O1). I guess the solution > is to switch to an autoconfigure script. The main reason for not > doing this is 1) time and actually more 2) the fact that the SDK does > not come with an ac script. > > Thanks for the feedback, Kasper <snipped>
I think I probably know about the breakage - gcc 4.x mis-compiles the Affymetrix toolkit at the default optimization (-02) shipped with the toolkit. (Prof Ripley's comment was spot-on - why the reluctance to just call it the Affymetrix toolkit bundled with the bioconductor package affyparser?). I narrowed it down to the strict-aliasing option... try '-O2 -fno-strict-aliasing' - it is not portable, but it works around the funny way some part of the affy toolkit (actually one particular file) is written. I did report it in the Affy forum and also directly to my Affymetrix dev contacts about a year ago; gcc 4 is the default on the mac these days, I think? On a related issue, I have some annoyance with cross-compiling leaving behind *.o files (and *.d, etc) which get in the way of native compile and vice versa, and I have been wondering if there is a recommended solution? What I am currently doing is to have a "Makefile.utils" (I don't want or need a proper Makefile), and do 'make -f Makefile.utils clean' when I need to clean up - it is better than typing rm -f *.o *.so *~ *.rc *.d *.dll Makedeps *.def (particularly with bash's file completion, all I am typing is 'make -f M<tab>f<tab> clean' to clean up). I am sure somebody will laugh at this... Hin-Tak ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
