Re: [PATCH] perf report: Add a simple GTK2-based 'perf report' browser

[Arnaldo Carvalho de Melo <acme@ghostprotocols.net>]

Em Thu, Feb 23, 2012 at 02:47:39PM -0200, Arnaldo Carvalho de Melo escreveu:
> Em Thu, Feb 23, 2012 at 06:45:04PM +0200, Pekka Enberg escreveu:
> > On Thu, 2012-02-23 at 14:39 -0200, Arnaldo Carvalho de Melo wrote:
> > > > Dunno, probably. Does that matter, though? It's pretty damn fast as-is.
> > > 
> > > Well, what kinds of perf.data file are you feeding it?
> > 
> > It's a real-world case where I'm profiling JRuby startup under Jato with
> > perf:
> > 
> > penberg@jaguar:~/src/jato$ ls -lh perf.data 
> > -rw------- 1 penberg penberg 453K 2012-02-23 18:06 perf.data
> > 
> > On Thu, 2012-02-23 at 14:39 -0200, Arnaldo Carvalho de Melo wrote:
> > > I'm testing it now with large ones, lets see.
> > 
> > How big files are we talking about here?
> 
> 5 MB ones, say.

Nah:

[root@felicio linux]# perf record -a -F 10000 sleep 5m
[ perf record: Woken up 166 times to write data ]
[ perf record: Captured and wrote 42.104 MB perf.data (~1839573 samples)
]
[root@felicio linux]# 
[root@felicio linux]# 
[root@felicio linux]# perf report --gtk

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND                                                                                                                           
10618 root      20   0  408m 170m  49m S  0.0  2.2   0:04.71 perf

[root@felicio linux]# perf report --tui

10633 root      20   0  379m 165m  45m S  0.0  2.1   0:01.15 perf
 
> But the way you did it is pretty minimalistic, which is good, its just
> that I don't really like the idea of having two mirror data structures
> representing the report lines.

- Arnaldo