This makes sure we are never going to doubly free it.

Warned-by: gcc ASAN
Stable-Candidate: 1.19.x

These strings are going to be auto-released on program exit.

Warned-by: gcc ASAN

This variable is usually set up once at the beginning of the execution
and then never changed again.

Warned-by: gcc ASAN

We need to free the last element from the tree.

Warned-by: gcc ASAN
Stable-Candidate: 1.19.x

Warned-by: gcc ASAN
Stable-Candidate: 1.19.x

Reported-by: KOLANICH <kolan_n@mail.ru>
Warned-by: gcc ASAN
Stable-Candidate: 1.19.x

We need to free the members when assigning into them for the GNU
longlink and longname extensions.

Warned-by: gcc ASAN
Stable-Candidate: 1.19.x

While varbuf_grow() handles 0 sizes fine, we do a check before calling
it to avoid the duplicate checks.

Reported-by: KOLANICH <kolan_n@mail.ru>
Warned-by: gcc ASAN

This makes spotting unexpected values, or output from sanitizers easier.

Unconditionally disabling the output makes diagnosing problems very
hard. Let's do that only on the default silent mode.

These are inconsequential, but shadow actual memory leaks from being
detected by address sanitizers or similar tools.

Warned-by: gcc ASAN

Check whether the database files exist and ignore them otherwise. If
some database file has been modified, but the current file being
processed does not exist, created an empty backup so that they are
all in sync with their sequence number.

Closes: #969472

When readlink fails inside a $() the whole script fails. Remove the
«set -e» from the bug-script, as making it fail will cause programs
like reportbug to make it way harder to report bugs, which is
counterproductive.

Closes: #968397

This got deprecated upstream in 2014, but only in 2020 the type started
to emit deprecation warnings.

Changelog: internal

Prompted-by: Niels Thykier <niels@thykier.net>

This is clear from the deb822(5) man page, but we can make this more
clear on the brief format description for each of these file formats.

Prompted-by: Niels Thykier <niels@thykier.net>

Changelog: silent

Prompted-by: Niels Thykier <niels@thykier.net>

Specify that the day-of-month accepts an optional leading zero, and
the amount of supported spaces.

Ref: #971977
Reported-by: Axel Beckert <abe@debian.org>

These have always been separated by a single space, in contrast to other
values in the file, but allowing more spaces now would break backwards
compatibility for something that's just aesthetics. Simply document the
expected format.

Closes: #970083

Makes it easier to see what this format is based on.

Prompted-by: #970083

Signed-off-by: Guillem Jover <guillem@debian.org>

Use uppercase for acronyms. Qualify GNU specific values with GNU.

Closes: #966111

We need to refer to --print-set instead of --print-unset.

Reported-by: Ferenc Wágner <wferi@debian.org>
Closes: #966110

The glibc project has deprecated this header for some time and made it
emit warnings, and removed it in 2.32. The Linux kernel has removed the
functionality behind the syscall in 5.5.

Warned-by: cpp

This is a requirement for the CPAN distributions.

We need to do the substitution before converting from POD to man,
otherwise the markup generated will make the regexes not match.

Reported-by: Niels Thykier <niels@thykier.net>

We need to extract the KEY_* macros from the curses header file that
we are going to end up including in the source. But we should not
assume any kind of structure, because that's just fragile, as the
contents can be included from other header files, or be protected
behind a pre-processor macro or similar.

Instead we use the CPP -dD option which will make the pre-processor
handle all the internal details and output a list of active macros
with their values, which we can then safely parse.

Closes: #970545

This tool will assist in preparing the debian/changelog out of the
git log, by using the specially formatted commit messages and meta
fields.

This option makes it possible to pass options to the item constructor
in new_item() method.

Changelog: silent

[guillem@debian.org: Hook into the build system. ]

Closes: #964751
Signed-off-by: Guillem Jover <guillem@debian.org>

Some of the options had not been explicitly initialized to their
implicit default value, which broke when the code to verify signatures
got refactored into Dpkg::OpenPGP::verify_signature() with a default
for require_valid_signature of 1.

This is required here too as this class does not call the initialization
from its parent. This is error prone, and will be refactored in the
future to make use of a common setup in the parent.

This way we do not require a dpkg(1) in the system, which will be common
on non-dpkg based systems.

On systems where the defaults for these programs are not the GNU
variants, we need to look for them and use them in preference. Search
in the CPAN distribution build system, for the names prefixed with «g»
which is the convention used on BSD systems for GNU variants, and use
them to replace the variables in the Dpkg module.