Hello!

Lintian in Debian unstable has a new test that my package is failing on: https://lintian.debian.org/tags/no-code-sections

Can somebody who recently fixed this in their own package give some
advice how I should fix it for mariadb-10.6?

See example of Lintian error about libmariadbd.a in context at https://salsa.debian.org/mariadb-team/mariadb-server/-/jobs/2203726


in https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=977596 there is a
shell example how to test for no-code-sections, but it does not return
anything for me locally:

readelf -W --section-headers ./debian/tmp/usr/lib/x86_64-linux-gnu/libmariadbd.a | sed -n '/^ *\[[ 0-9]*\]/s/ *\[[ 0-9]*\] *//p' | awk 'BEGIN {rv=1} /^NULL/ {next} $1 ~/^.(text|data|(preinit|init|fini)_array$)/ {if ($5 !~ /^0+$/) rv=0}
END { exit rv}'
readelf: Warning: unable to apply unsupported reloc type 21 to section .debug_info

(or https://paste.debian.net/1220493/ with no line breaks)

Any suggestions how to debug this locally so I can validate if various
changes in debian/rules fixes it or not?


- Otto

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Fri, 26 Nov 2021 at 22:10:45 -0800, Otto Kekäläinen wrote:

Lintian in Debian unstable has a new test that my package is failing on: https://lintian.debian.org/tags/no-code-sections

As with any Lintian tag, I would recommend doing your own assessment
of whether this is a real error before assuming there is a bug that you
need to fix.

This particular one looks like it's a Lintian bug: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1000469 (fixed in git)

Looking at the results reported with current Lintian, it seems to me
that this check was initially assuming that every object file (.o) in
a static library is expected to contain at least some executable code,
but I don't think that's really true:

* some translation units end up effectively empty, because they contain
only functions that are #ifdef'd out under the build options in use,
for example SDL_evdev_kbd_freebsd.o in a Linux build of libSDL2.a

* some translation units contain global variables but no executable code,
for example (guessing from its name) decomposition-table.o in libunistring

The resolution in Lintian git was to make this check only trigger if
*the entire static library* contains no executable code, which seems
reasonable (and in the rare case that a static library intentionally
contains only variables and no code, it would be appropriate for a
maintainer who knows that to override the hint).

If feasible, an excellent way to verify that a static library is
practically usable would be to add an autopkgtest that statically links
a simple executable that calls one of your library's functions, and then
runs that executable. If you add a test like this, please remember to
flag it as 'superficial' to indicate that it doesn't provide thorough
test coverage.

smcv

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Sat, Nov 27, 2021 at 5:30 AM Simon McVittie <smcv@debian.org> wrote:

On Fri, 26 Nov 2021 at 22:10:45 -0800, Otto Kekäläinen wrote:

Lintian in Debian unstable has a new test that my package is failing on: https://lintian.debian.org/tags/no-code-sections

As with any Lintian tag, I would recommend doing your own assessment
of whether this is a real error before assuming there is a bug that you
need to fix.

I tried, but clearly reading the Lintian page linked above and the bug
report linked from it didn't really convey what it was actually trying
to achieve. Now when you explained it, I got it, and also see why it
is fine to ignore.

This particular one looks like it's a Lintian bug: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1000469 (fixed in git)

Thanks for pointing this out!

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)