On Thursday, 4 December 2014 at 10:16:52 UTC, Martin Nowak wrote:
On Thursday, 4 December 2014 at 08:14:56 UTC, Jacob Carlborg
hack. I have stopped following the std.log thread, could you
summarize what issues you're having and trying to solve and
why not a more obvious solution doesn't work?
Zero overhead for statically disabled log levels.
The current proposal uses global version identifiers to alter
the behavior of std.logger.
Fine grained (at least per library) control over log levels.
The current proposal required to use different logger wrappers
in each library and scoped global version identifiers, see [1]
in the OP.
I like the direction you are taking, but I think the better
solution is to have:
1. A manifest/config file where you configure these and other
settings in a uniform manner.
2. Provide support for configurations and switching between
configurations at realtime.
3. Compiler support for the manifest file that allows library
code to query it at compile time as well as generating bitfields
for real time switching between configurations. (Just memcpy to
change configuration)
4. Have logInfo(…) etc as always inlined stubs.
5. Have one bit per log level (32 should be sufficient), rather
than a number.
6. Allow for compiler profiling so that bits in the configuration
that are accessed together become co-located on the same word/64
bytes cacheline.
Something like this:
<configuration name="normal">
<module match="*">
<logging type="info" status="off"/>
<logging type="error" status="on" logger="file"
href="file://var/myfile.txt"/>
</module>
<module match="http">
<logging type="info" status="on" logger="remote"/>
</module>
</configuration>
<configuration name="debug" inherit="normal">
<module match="mylib.*[has_symbol('DEBUG')]">
<logging type="debug" status="on" />
</module>
</configuration>
