In systemd-233 the format of the constants file changed to use
SD_ID128_MAKE_STR macro and long lines are broken with '\'.
Doing this in sed is too anyoing — add a simple python script to
do the processing.
send() already does conversions in a type-specific way, and doing it
in journal handler would defeat those conversions. In particular, UUIDs
would be converted to early and have dashes.
Removed mid variable from JournalHandler since the MESSAGE_ID is already
in the extras variable. MESSAGE_ID was being set to None, but this won't
appear in the logs.
This way we can skip ENOENT (which happens in containers).
While at it, let's extend the tests a bit, so that we at least
call all functions and check the type of the return value.
Also, drop '.nspawn' from the machine name, nspawn doesn't
use that suffix any more.
In _reader we were raising OSError(errno=ENOSYS), but in _dameon
we were raising NotImplementedError. Let's always use ENOSYS.
Also, make the messages in _reader more specific.
Fixes#33.
When running in a chroot, doctests that called journal.stream would
fail with ENOENT. Move the tests to test_journal, where we can skip
tests properly (without uglyfying the documentation).
Fixes#32.
It's fairly easy to provide a reasonable default for the first argument.
Let's do that.
Also, the documentation was misleading, suggesting that the function itself
can be passed as file parameter to print(). Use a different name for the
temporary variable to clarify that.
Instead of generating the list of message ids anew during every build,
the file is generated manually and committed into the
repository. Also, the list of defines is stored in id128-defines.h,
also kept in the repository. Both files should only grow.
This should make build easier. But it also fixes a problem with
systemd, which occasionally drops message definitions. We will keep
them forever, so it should be safe to rely on the presence of message
definitions which systemd does not use anymore.
Fixes#23.
This somewhat breaks backwards compatibility, but not for the
previously documented arguments: floats are now interpreted
differently, but ints and datetime.datetime objects are interpreted
the same as before. But the documentation clearly stated that only
ints and datetime.datetime objects were allowed.
This makes seek_realtime match seek_monotonic and other functions
which take time and follows the principle of least surprise.
Fixes#21.
Let's pass any flags through to the journal functions without checking
validity. Those functions do their own checking, so there's no need to
second-guess.
The semantics for _Reader(flags=0) are a bit changed:
before, this would be transformed into sd_journal_open(SD_J_LOCAL_ONLY).
Now, this results in sd_journal_open(0). Previous behaviour should be
achieved by not specifying the flags at all.
This change is necessary, because previously it was not possible to
pass flags=0 to sd_journal_open(0), i.e. it was not possible to "merge"
journals through the Python interface.
Similarly, Reader(flags=0) now means to open all journals, and
Reader(flags=None) is the same as Reader(flags=LOCAL_ONLY).
Wrapping the sources to ~80 columns means that the formatted output is
annoying to read. Rewrap to ~74 columns in the output.
Also remove some obsolete descritions of journal permissions and refer
to journalctl(1) instead.
Add some missing docstrings.
gcc warns that r might be uninitialized, because it doesn't
know that r will be initialized in the 'if' statement.
Initialize the variable to avoid the warning.
