Mike Hearn wrote:
By the way, I talked to Alexandre about the GCOV stuff, basically he wants us to find out why arc profiling borks the preloader. Then the idea is it can be enabled just using CFLAGS.
(You're confusing the UCLA guys with the Harvey Mudd guys, Mike!)
Here's what aarvey@cs.hmc.edu wrote when he ran into the problem:
make[1]: Entering directory `/home/aarvey/winefromwinehqcvs/wine.patched/loader' gcc -o wine-preloader -static -nostartfiles -nodefaultlibs -Wl,-Ttext=0x78000000 preloader.o -L../libs/port -lwine_port preloader.o(.text+0x1929): In function `_GLOBAL__I_wld_startGCOV': /home/aarvey/winefromwinehqcvs/wine.patched/loader/preloader.c:955: undefined reference to `__bb_init_func' collect2: ld returned 1 exit status make[1]: *** [wine-preloader] Error 1 make[1]: Leaving directory `/home/aarvey/winefromwinehqcvs/wine.patched/loader' make: *** [loader] Error 2
I can compile the preloader just dandy as long as I get rid of the extra compile flags (-ftest-coverage and -fprofile-arcs).
You could provide a dummy implementation of __bb_init_func in the preloader, but that's uglier IMHO than just not passing those options when compiling the preloader.
Mike also asked:
Until that happens though are you generating lcov output anywhere?
Not yet, but I'll ask them to start doing periodic LCOV runs and putting them up on the web. I'd like to see this be a regular feature at winehq.org, not sure what the logistics would be. - Dan
Dan Kegel dank@kegel.com writes:
You could provide a dummy implementation of __bb_init_func in the preloader, but that's uglier IMHO than just not passing those options when compiling the preloader.
Sure, but you have to do that in a way that doesn't depend on configure, and doesn't require running sed magic on the command-line options.
-
Alexandre Julliard -
Dan Kegel