In chapter 7 we do not need to unmount file systems to strip binaries.
Move unmounting to the backup/restore sections.
Also make sure the stripping commands are valid.
Without them, for example:
cp /usr/lib/libbfd.so /tmp/libbfd.so
# now /tmp/libbfd.so is a hardcopy of /usr/lib/libbfd-2.37.so
strip --strip-unneeded /tmp/libbfd.so
install -vm755 /tmp/libbfd.so /usr/lib/libbfd.so
# now /usr/lib/libbfd.so is *not* a symlink, but a real file
We don't want this to happen.
libdl and libpthread removed: they are now dummy libs and is not used by
anything built in LFS/BLFS.
Update to GCC-11.2.0 (#4883)
Update to inetutils-2.1 (#4892)
Update to automake-1.16.4 (#4894)
SHA256 checksum entities for the three packages are added. I think we
can start a transition to SHA256 now.
The behavior of --strip-unneeded and --strip-all are same for
executables and shared libraries. For static libraries,
--strip-unneeded keeps the symbols needed for relocation processing so
it won't break static archives.
Move coreutils library from /usr/libexec to /usr/lib in both
Chapter 6 and Chapter 8.
Text updates in chapter08/pkgmgt.xml and chapter09/locale.xml.
Restore deletion of hanging test in chapter08/util-linux.xml.
Text changes in stripping and make instruction compatible with jhalfs.
Glibc loads these "NSS modules" (see man 5 nss) on startup. This is
implemented with something like dlopen() so "ldd" won't show them up.
So they should be considered "online" here.
Without this addition, overwriting libnss_files-2.33.so reproducibly
triggers a crash during jhalfs runs where stripping is enabled. In
manual builds, it reproducibly triggers a crash exiting from chroot.
After this change I reran stripping 5 times and there was no crash
observed.
In stripping, /usr/bin/bash, /usr/bin/find, and /usr/bin/strip are
running. Strip them, and all libraries used by them in /tmp, then
install them back.
We can't use this for all libraries or binaries: the process above
discouples hard links (for example /usr/bin/perl and perl5.34.0). So
unfortunately the stripping instruction is now a stupidly long bash
script...
For the minimum I have used allnoconfig, rounded down to allow for
future removal of redundant items (this is unlikely to produce a
bootable kernel), with its SBU rounded down for people with fast
machiens who build on NVMe or in RAM.
For maximum and typical, space is rounded.
Use a fixed UID for "tester" user, instead of the tty owner's UID.
The problem using tty owner's UID is, when tty is owned by "root" the
UID of "tester" will be 0. This cause a namesis between "tester" and
root. And, the "tester" is introduced to run some tests which are not
expected to be ran as root. If "tester" has UID 0, it will have root
priviledge and the tests will fail again.
Bash tests require the user to own the tty. To satisify it, we use
Expect to spawn a new pseudoterminal to run Bash tests.
The default timeout is only 10 seconds, where the bash tests won't
complete on most systems. Then expect will exit early, with bash tests
running in background. We won't like this.
