While this is true, Mark, the existing trace "packages" aren't particularly effective either, IMHO, for the following reasons:
* They are completely inconsistent (see opt_range.cc vs. storage/myisam vs. Innodb) * They are hacked into the server using a metric ton of #ifdef's which cause build and test failures out the wazoo * They are integrated into the code files, instead of being ancillary (like dtrace is...) I believe the current plan is to get rid of the DBUG library (include/my_dbug.h) and the 20K+ lines of DBUG_XXX code, and then to backport custom tracing functionality (for a good example, see the dbug_print, print_quick, etc, code in opt_range.cc) into a separate library that can be called via a debugger or other client program. Would this work for you? -jay MARK CALLAGHAN wrote: > gdb isn't a replacment for a trace package. If you remove DBUG and > don't add back trace statements via another package, then it will be > much more expensive to diagnose misbehavior. Debuggers are great when > you know what you are looking for. They are inefficient when you don't > know what to look for. If you only work on a small portion of the > code, a debugger is OK. If you have to work on code throughout the > server, then not having tracing is awful. And I think that anyone > doing support (yes, a lot of the time, I am doing support) will agree > with me. > > And if you make the trace package nice enough, you might be able to > get storage engines to use it. > _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
