The ebuild for smarteiffel-2.2 seems to compile an installation script of sorts, and then echo preselected values into that program. The program is install.bin, compiled from install.c. When I customise the ebuild and remove the pre-echoed values, in order to manually walk through the install script, it fails with a signal 11 on the first possible keystroke (it asks you to hit enter, and fails immediately).
Created attachment 128829 [details] The build log.
Created attachment 128831 [details] A diff to apply to the stock ebuild. This simply produces an alternate ebuild that should let you walk through install.bin manually (this is what I used to find exactly where the segfault was occuring).
Created attachment 128833 [details] My emerge --info. Note that I tried to remove all CFLAGS to test this, but apparently emerge reapplies -O2 and -pipe.
Interactive ebuilds are really not acceptable, I guess the patch is meant for debugging only? :)
(In reply to comment #4) > Interactive ebuilds are really not acceptable, I guess the patch is meant for > debugging only? :) > Sure :-). It seems a somewhat clumsy system to echo values into a script, but it works... or should, at least.
Thanks for the report. This is not specific to amd64, but it is (so far) specific to gcc 4.2, specifically with the -fstrict-overflow option which gets enabled at -O2 and higher. smarteiffel generates not completely valid C code and relies on the compiler not to optimise too aggressively, but gcc 4.2 does. I've updated the ebuild to add the -fno-strict-overflow compiler flag when supported to work around this issue, and I'll inform upstream. Hopefully, they will fix the code so that this will not be necessary in the next version. I'll close this as UPSTREAM instead of FIXED because really, this flag just hides the error.
