Running within the build sandbox has a number of disadvantages:
- We have a separate clangd cache for each distribution/release combo
- It requires to build the full image before clangd can be used
- It breaks every time the image becomes out of date and requires a
rebuild
- We can't look at system headers as we don't have the knowledge to map
them from inside the build sandbox to the corresponding path on the host
Instead, let's have mkosi.clangd run clangd within the tools tree. We
already require building systemd for both the host and the target anyway,
and all the dependencies to build systemd are installed in the tools tree
already for that, as well as clangd since it's installed together with the
other clang tooling we install in the tools tree. Unlike the previous approach,
this approach only requires the mkosi tools tree to be built upfront, which has
a much higher chance of not invalidating its cache. We can also trivially map
system header lookups from within the sandbox to the path within mkosi.tools
on the host so that starts working as well.
Let's add a basic clang-tidy check to the linter workflow. This
gives us the following:
- A check so that we don't introduce any new cyclic header dependencies
- A check to make sure all of our header files are standalone, as clang-tidy
will fail to parse header files that don't include all their dependencies.
The clangd include checker is rather broken and is littered with false
positives in our codebase, so let's disable the feature until it improves
a bit.
We already have LOG_CONTEXT_PUSH_EXEC() which with two additions
does exactly the same as the custom logging macros, so let's get rid
of the custom logging macros and use LOG_CONTEXT_PUSH_EXEC() instead.
Additionally, using the log context makes sure the extra fields are
applied to all log messages generated while the context is in place,
rather than only log messages logged with log_unit_xxx() in timedated
itself.
It also means the unit name is prefixed to all log messages logged
within that context. While it's not clear whether we always want the
unit name to be attached to library log messages, we also don't want
the unit name to not be attached to any library log messages, so we opt
to add more rather than less information by adding the unit name everywhere
for now.
In the future we can look into some log.h helpers to enable/disable
adding the prefix to the following log messages or not.
Currently LogLevelMax= can only be used to decrease the maximum log level
for a unit but not to increase it. Let's make sure the latter works as
well, so LogLevelMax=debug can be used to enable debug logging for specific
units without enabling debug logging globally.
ASSERT_PTR() and friends in assert-fundamental.h make use of assert()
and assert_se() which when not building for sd-boot are defined in
assert-util.h. Because assert() from glibc is only overridden in
assert-util.h, the macros in assert-fundamental.h still end up using
the glibc assert.
Let's fix this by moving the required macros and related logic to
assert-fundamental.h.
assert() is compiled away if NDEBUG is set which causes an unused
variable warning in various places when the next commit is applied
so let's use assert_se() to avoid these warnings.
Tooling such as clang-tidy is bad at dealing with condition
compilation on the build system level instead of at the source file
level. What happens? It still tries to analyze the file and fails
horribly if the required headers aren't available. Let's work around
the issue and make things more consistent at the same time by doing
the necessary HAVE_GCRYPT checks inside of the source files instead
of doing them at the build system level.
We also add some typedefs to allow getting rid of various HAVE_GCRYPT
checks.
This check doesn't work properly when tools such as clang-tidy are
analyzing headers. Since we don't care about the check in that case,
skip it if __clang_analyzer__ is defined similar to how we skip it as
well if __COVERITY__ is defined.
The static inline declaration in iovec-fundamental.h causes static
analyze tooling warnings (clang-tidy). Let's get around it by making
free() non-inline. LTO will make sure it's still inlined properly.
We want to make the header standalone so it includes all the stuff it
needs. However, including macro.h for assert_cc() doesn't work because
of generate-audit_type-list.sh which would have to become more complex
to handle the extra include directories.
Instead, let's just use _Static_assert() directly which is a builtin and
doesn't need any extra includes.
These depend on a full definition of DnsResourceRecord which we want
to forward declare in the next commit, so let's move the implementation
of these functions to the implementation file and make them lowercase.
Include the headers in the implementation file (and the gperf file)
instead of in the qdisc.h header. I opted to keep them separate from
the other headers since they're slightly different than regular headers.
Currently there are various circular dependencies between headers
in core/. Let's get rid of these by making judicious use of forward
declarations and moving includes into implementation files instead of
having them in header files.
Getting rid of circular header includes simplifies the code and makes
various clang based tooling such as iwyu work much better on our code.
The most important change is getting rid of the manager.h include in
unit.h which is possible thanks to the previous commits. We also move
the OOMPolicy and StatusType enums to unit.h to remove the need for
other unit headers to include manager.h to get access to these enums.
There's no need for these to be fields inside the manager struct,
let's turn them into functions in unit.h instead, again to allow
forward declaring the Manager struct in a later commit.
For example, consider the following device:
- syspath: /sys/bus/mdio_bus/drivers/Qualcomm Atheros AR8031!AR8033
- subsystem: "drivers"
- driver subsystem: "mdio_bus"
- sysname: "Qualcomm Atheros AR8031/AR8033" <-- '!' is replaced with '/'
When sd_device_new_from_subsystem_sysname() is called to get the device,
the arguments to sd_device_new_from_subsystem_sysname() should be
- subsystem: "drivers"
- sysname (concatenated with driver subsystem): "mdio_bus:Qualcomm Atheros AR8031/AR8033"
In that case, we need to pass to device_new_from_path_join() the following:
- subsystem: "drivers"
- driver subsystem: "mdio_bus"
- sysname: "Qualcomm Atheros AR8031/AR8033"
- a: "/sys/bus"
- b: "drivers"
- c: "mdio_bus"
- d: "Qualcomm Atheros AR8031!AR8033"
Here, the important point is that the `sysname` argument and the
last argument `d` are differnt: the `sysname` argument needs to match
the sysname obtained by `sd_device_get_sysname()`, but `d` must be
the last path component of the syspath.
Previously, we passed a wrong sysname to device_new_from_path_join().
This fixes the issue.
Fixes a bug in cd7c71154c (v257).
I needed something to hack on "offline", during a long transatlantic
flight. I decided to implement #35862, since I could do that without
checking any online docs nor needing GitHub.
Daan De Meyer
cvlc12
Yu Watanabe
Mike Yuan
Lennart Poettering