On Tue, Aug 13, 2013 at 10:44:07PM +0100, Ken Moffat wrote: > On Sun, Aug 11, 2013 at 09:14:54PM -0500, Bruce Dubbs wrote: > > > > Try adding a namespace, e.g. KM_SCRIPTS, to your environment variables. > > However, I suspect a problem in some package earlier in the build. > > > Thanks for that comment - I didn't understand it properly at the > time (unsetting SCRIPTS in automake works for me), but I'm now using > a debug (git) branch of my scripts and I've discovered that my bison > build breakage happened between 2.7.1 and 2.7.90. So far, I'm only > 16000 lines through the diff (4%), but I now understand that > everything outside the package itself needs a namespace for its > variables. Will probably take me the rest of the night to change > them all
tl;dr - sorted, by using a namespace throughout my build. Well, that was a _severe_ underestimate. More like "rest of the week". Changed almost every variable in my scripts themselves, and my functions. Then changed the function names. Didn't help. Spent more time looking at that bison diff, but it's too big to manage. My guess is still on something in gnulib (or perhaps automake). Tried using git bisect, but couldn't get any random changeset to build - there are submodules (gnulib, automake) and I'm not sure that they were being pulled as the matching versions. Perhaps the source is just like my own scripts, and often doesn't build. Merged my changes back to my trunk, fixed breakages - that took about a day in itself. Anyway, put in an ugly workaround to untar bison and stop, and then to test if the bison binary existed when trying to continue. That let me continue, but the space and time measurements for bison were rubbish and I hated not being able to leave it running. While I was fixing up the merges, I realised that I'd overlooked one set of variables - I use a 'packages' file to list package versions and patches, so it contains things like BISON=bison3.0. Spent some hours today changing ALL of those lines, and a few random other variables I'd overlooked in my chroot driver script such as ELFINT which contained the Requesting program interpreter details for different architectures). One of those variables (about 75 of them if you count commented-out ppc stuff) did the trick. Leaving it to run, but I'll be back to testing 7.4 (if I don't go away). Almost week wasted. <sigh/> ĸen -- das eine Mal als Tragödie, dieses Mal als Farce -- http://linuxfromscratch.org/mailman/listinfo/lfs-support FAQ: http://www.linuxfromscratch.org/lfs/faq.html Unsubscribe: See the above information page