I’ll note that MessageSupplier is deprecated, and we eventually added support
for Supplier<Message>, so that’s already there. Since it sounds like we have
some good ideas on what a minimal v3.Logger API would look like, it would be
cool if anyone would like to propose a full API. Here’s another Logger API to
consider (from OpenTelemetry, something I’ve been considering for future
integration purposes, though this might be mostly on the Flume side of things):
https://javadoc.io/doc/io.opentelemetry/opentelemetry-api-logs/latest/io/opentelemetry/api/logs/LogRecordBuilder.html
(their Logger returns this builder class).
—
Matt Sicker
> On Nov 23, 2023, at 03:31, Piotr P. Karwasz <piotr.karw...@gmail.com> wrote:
>
> Hi Matt,
>
> On Tue, 21 Nov 2023 at 23:22, Matt Sicker <m...@musigma.org> wrote:
>>
>> This sounds like it might be a good basis for figuring out a parallel v3 API
>> for a “hard to mis-use” style API. However, once you go that route, you
>> start to wonder how useful templated log messages are when you can capture a
>> lambda instead. Parameterized log messages might work better as structured
>> log messages, something that is awkward to use in the API at the moment.
>
> If we'll create a separate `v3.Logger` interface I would clean it up
> from many methods, e.g.:
>
> * getLevel() and getName(): how are these useful for the user? An `if
> (logger.getLevel() == Level.INFO)` should be replaced by `isEnabled`,
> * getMessageFactory() and getFlowMessageFactory() (the latter is my
> fault): again these are not useful to the user. If I need a message
> factory, it will be a different message factory,
> * printf(): a better approach is to use StringFormatterMessageFactory
> for the logger,
> * catching(Throwable): can be replaced with `error(Object)` and the
> semantics described in this thread,
> * throwing: does anybody use it? Maybe it could stay,
> * entry/exit, traceEntry/traceExit: I can not imagine using these on
> each method (or important method). I'd rather use AspectJ pointcuts
> instead (or a @LogTrace annotation),
> * methods that use `MessageSupplier` like `info(MessageSupplier)`:
> couldn't these be integrated into the logic of `info(Supplier)`?
> * the `is*Enabled` methods are prone to misuse: a snippet like:
>
> if (logger.isDebugEnabled()) {
> logger.debug(MARKER, "Hello world!");
> }
>
> will not print any message if the level of the logger is less specific
> than DEBUG, even if the user asks for **all** MARKER messages to be
> printed.
>
> IMHO opinion `v3.AbstractLogger` should only have 2 abstract methods:
> * logMessage(Level level, Marker marker, String fqcn,
> StackTraceElement location, Message message, Throwable throwable)
> * isEnabled(Level level, Marker marker, String fqcn,
> StackTraceElement location, Message message, Throwable throwable)
>
> Piotr
