b15b5c9a11
Highlight: inlining on function trees should be active in C++. Reword include files fixing so that it wedges in where it is supposed to, as latest tweaks to va_list means the stage1 compiler needs correct includes. |
||
---|---|---|
.. | ||
alpha.readme | ||
i386.readme | ||
m68k.readme | ||
README | ||
sparc.readme |
testing procedure: - obtain a RECENT snapshot of dejagnu from egcs.cygnus.com. I've just upgraded to dejagnu-981026, check ftp://egcs.cygnus.com/pub/egcs/infrastructure. It's dreadfully easy to build it, just a trivial adaptation of the existing dejagnu port. - make build should now BOTH build the snapshot AND build the checks. You will also find a `warnings' file in the work/egcs-* directory... - have a look at work/source/contrib... this is what you need to analyze the test results. OpenBSD notes: * you need an up-to-date bsd.port.mk. * an assembler bug that affects the sparc has been fixed post-2.4. It may affect other architectures. Basically, the archs that use an old gas (m68k, sparc, i386,...) are liable to run into bugs that can be traced to the assembler, not the compiler. * the configuration files used are substantially cleaned up from gcc 2.8.1. This will be used for gcc 2.8.1 eventually. To run the same testsuites under native gcc 2.8.1: - move the site.exp file elsewhere, edit it to get the right paths. - remove the -Os option from the various testsuite drivers, gcc 2.8.1 does not support -Os... - run with e.g., GCC=/usr/bin/gcc runtest --tool gcc