On 2/6/08, Attila Lendvai <attila.lendvai@...> wrote:
> dear list,
> please find attached an updated version of the frame cleaning patch
> already discussed in this thread (there are only comment changes
> relative to the one i've sent previously).
> the second patch factors out internal MAP-BACKTRACE and MAP-FRAME-ARGS
> functions that came handy when working on the deadlock detector (which
> is still lingering in some random files here looking for a home
> together with the PATH-TO-ROOT that helps tracking why stuff is not
> gc'd. would sb-introspect be an appropriate home for them?)
The refactoring patch looks good -- I'm merged as 184.108.40.206, thank you!
I still have some issues with the method frame cleaning:
While FAST-METHOD, XEPs, etc, are not of interest of the "average
hacker" -- not even most SBCL hackers most of the time, I suspect --
PRINT-FRAME-CALL :VERBOSITY also controls source printing, which is
(if you like that sort of thing) a potentially user-friendly feature.
Conflating "useless internals noise" and "show the source for this
frame" is not good.
If also doesn't deal with SB-PCL::SLOW-METHOD frames -- which while
rare, also exist. Is this by design?
So, I'm thinking that make *VERBOSITY* mean just "internals noise",
and add a :PRINT-SOURCE argument to PRINT-FRAME-CALL.
1: default -- maximally cleaned frames
2: don't clean actual function names (right now just PCL stuff)
3: don't clean entry point details
How does this sound?