Age | Commit message (Collapse) | Author |
|
Instead of computing the shard based on the RemoteExecutionConfig
singleton, use the already computed hash stored in the passed
StorageConfig instance, which now needs to be set up separately
if bootstrapping in order to avoid unwanted includes.
Storing the backend description to CAS and corresponding audit
checks now take place in main.
|
|
...such that it will be available to the TargetCache for sharing.
Also, GC does not require remote execution information, so the
logic for this subcommand is moved earlier in main.
|
|
...and replace it with passed instances created early via a builder
pattern.
|
|
|
|
|
|
As the storage instance also instantiates the target cache, we need
to ensure that each such instance has at that point all the correct
remote endpoint information to ensure the target cache sharding
works as expected.
In particular, the server-side and client-side operations of just
have a slightly different setup for the remote endpoint address,
therefore they require the storage instantiation to be done
separately.
|
|
|
|
|
|
|
|
...and adjust AnalyseContext.
|
|
|
|
|
|
|
|
|
|
|
|
...to track changes during refactoring easier.
|
|
Use a builder pattern for creation and validation, in a manner that
allows also other authentication methods to be added in the future
besides the current TLS/SSL.
The main Auth instances are built early and then passed by not_null
const pointers, to avoid passing temporaries, replacing the previous
Auth::TLS instances passed by simple nullable const pointers. Where
needed, these passed Auth instances are also stored, by const ref.
Tests also build Auth instances as needed, either with the default
'no certification' or from the test environment arguments.
|
|
|
|
...to more easily keep track of changes during refactoring.
|
|
|
|
|
|
|
|
...instead of std::optional<gsl::not_null<RepositoryConfig const*>>.
|
|
The build tool prints status messages at the end of each of the
major stages that sequentially follow each other. This allows,
in particular when using log files (that contain time stamps) to
quickly identify which phase was the resource consuming one. All
these message mention the (same!) configured target that is process
by this invocation of the build tool. In the presence of toolchains,
the configuration can, however, get quite large (containing lots of
paths to various tools). Therefore, shorted the configuration in the
repeated menions if it is very long (currenlty 320 characters); as
we do not change the first mentinoing ("Requested target is ..."),
even with that shortening the log contains the same information.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
...and adjust interfaces.
|
|
...instead of initialization of the singleton.
Use builder for creation and validity check.
|
|
...and adjust interfaces.
|
|
...instead of using singleton calls.
|
|
...instead of using singleton calls.
|
|
...instead of using singleton calls.
|
|
...instead of using singleton calls.
|
|
|
|
...to track changes during refactoring easier.
|
|
... for failures instead of the binary strict_logging. In this
way, we can log at log level Error in the serve instance, and for
the client at level Warning or Info, depending on whether failure
is expected (i.e., if targets are potentially taken from a serve
endpoint) or not.
|
|
Once a RepositoryConfig instance gets populated, it must never be
changed again. Therefore, all functions accepting these instances
should only take them as pointers to const.
|
|
... to simplify reuse.
|
|
... so that they are available in machine-readable form. In this
way, all logs can automatically be collected without the need of
parsing human-targeted error messages.
|
|
|
|
|
|
... so also report at info level if we got export targets served.
|
|
|
|
|
|
...in the tool proper, i.e., excepting tests.
|
|
Some of the more specific issues addressed:
- missing log_level target/include
- header-only libs wrongly marking deps as private
- missing/misplaced gsl includes
|