Commit
16abb5971ff3430d84ae62b139208c5b1531080d
by Helmut Grohne
build make-dfsg with a noguile profile
Yeah, this gets us more profiles, but at the same time, it likely drops all guile from the build and with that repeated patching for every new architecture and it saves a bit of build time.
Commit
b80a6b900f743582c3ebb9bd29494959c2b9e66c
by Helmut Grohne
improve usage of dpkg-architecture
* Properly quote arguments, which also makes shellcheck a little happier. * Use the -i flag rather than matching its output. * When in a hook, prefer the hook argument as architecture.
Commit
7b6b07bfa607abea2da0f7aaeb34c758c03e4f37
by Helmut Grohne
unconditionally apply patch_gcc_has_include_next
gcc-12 builds started failing in stage1 while compiling libgcov-merge.c due to missing <sys/mman.h>. That header is only available in glibc stage2, so the issue is that HAVE_SYS_MMAN_H is defined. That happens, because the relevant configure check erroneously succeeds by using /lib/cpp rather than the correct cpp invocation. cpp is misdetected, because the correct cpp failed #including <limits.h>, because our patched __has_include_next(<limits.h>) produces an error, which is gcc bug 80755 and is fixed by this very patch. A subtle change in include directory order made this patch important and the failure came three layers later.
Commit
cacf3bf31cd1ecffb4e44db9e4a902242964c8a6
by Helmut Grohne
increase shellcheck happiness
These are relatively obvious fixes that e.g. improve compatibility with bash, add missing quoting or use idiomatic computations.
What remains for shellcheck largely is style remarks and false positives. For instance, we excessively modify variables in subshells to intentionally get those changes reverted. Likewise, using variables in single-quoted strings is not a mistake. In order to address the globbing issues, more architectural changes would be necessary.
![[Jenkins]](/static/39869d34/images/svgs/logo.svg){kind=logo}
