[Mono-dev] New profiler is "ready"
paul at science.uva.nl
Fri Mar 28 06:51:04 EDT 2008
Is the new profiler supposed to replace tools like heap-buddy and
heap-shot or is it a supplement to them?
And is the statistical profiler similar to what oprofile can offer?
Massimiliano Mantione wrote:
>the new logging profiler is now usable on Linux, and I'd like
>some feedback and bug reports.
>It is described here:
>and in the mono.1 man page in svn.
>Open issues (for now):
>- The default name of the output file should contain the
> application name, and not just be "profiler-log.prof".
>- The allocation profiler still attributes object creations
> to wrappers instead of the caller method.
>- I should check why the decoder is not properly installed
> and also change its name to something more usable.
>- The statistical profiler tries to get the names of
> unmanaged functions using dladdr, but if this fails, it
> has an "oprofile like" behavior: it saves the IP value
> and at *decoding* time examines the elf files (using nm)
> looking for the symbol name.
> This is a bit sophisticated (it looks at all the memory
> regions of the process using /proc/self/maps at run time
> so that the correct file is examined at decoding time), and
> works fairly well (better than dladdr).
> However, this is fragile: if the decoding time happens
> when some of the relevant elf files has changed, the info
> is lost and the data in the log file is useless (and there
> is no warning of this).
> I am moving code from the decoder to the profiler, so that
> the symbols are all looked up at runtime, but this code is
> not ready yet.
>- The heap analysis features technically work, but are not
> yet exposed to the user by the decoder program.
>That's it... have fun :-)
>Mono-devel-list mailing list
>Mono-devel-list at lists.ximian.com
More information about the Mono-devel-list