I got all the way through chapter 6 with only a few hang ups. Sed compiled with sed-4.1.5-fixes-1.patch builds cleanly, passes all tests, but has very odd behavior. At first, it could not operate on files that were not in the same directory if you used the -i.orig option. It errored out with "Cannot rename directory. Directly not empty." or something of that nature unless you used the -e switch as well. I somehow lost that bug in the process of recompiling and installing again, but even after that. If you used the -i.orig option, it would do the substitution correctly, but you would end up with an excess file. Say you tried sed 's/test/pass/' -i.orig test/try.1 that only contained the word test. It would do the expected substitution, but there would be a new file called tes in the current directory that would be a duplicate of test/try.1 that also said pass. I didn't really like the buggy nature and reinstalled without the patch.
gmp and mpfr are supposedly required for building gcc. It actually builds without it, but I decided to go ahead and install those packages prior to the gcc build. You cannot run the gcc testsuite without autogen which requires guile which requires mpfr and gmp. I finally got autogen working, but make sure you build guile with thread support or autogen won't even compile. I didn't manage to figure everything out until automake, autoconf and libtool were all installed in chapter 6, so I didn't get a chance to run the gcc testsuite anyway. perl claimed gethostbyaddr didn't work. I did not have time to research this and it was the only perl test that failed. I did have /etc/hosts with 127.0.0.1 localhost livecd udev had to be updated. I used udev-105 I used the most recent kernel 2.6.19.2 and had to update the grsecurity patch for it. After that, I got stuck because of the incorrect assert defines described elsewhere in this list. :) Starting over again. This time I will get the gcc test working so that I can actually test it before the install. I had a lot of packages that I couldn't compile with -Werror enabled because of warn_unused_result attributes for write, printf, scanf, etc. There was one other patch that I had to update, but I forget for what. It's in the LFS patches archive. When compiling development books, should remind myself to always search the patch archive for newer patches. I'm going to write down the steps this time because I pretty much have an idea of what works now. Would anyone be interested in a detailed posting? I'm just wondering if compiling the toolchain as one unit has ever worked for anyone. If it hasn't, maybe the structure should go back to headers-binutils-gcc again? Charity -- http://linuxfromscratch.org/mailman/listinfo/hlfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
