[bind10-dev] discussion about the approach of gprof

[zhanghaikuo]

Jinmei wrote:
>Now, the possibly related question to the original point: do we still
>have the same-output-file problem if run it within the source tree?
>If not, we'll not have to use GMON_OUT_PREFIX, much less modifying the
>bind10 script.  We might still mention the tweak to the bind10 script
>in case someone really wants to run it under an installed environment,
>but I guess it would be a supplemental note.

My answer is yes. the "same-output-file problem" exists there if we run it in the source tree.
because the gmon.out will be produced in the working directory(the directory which we run bind10 scripts),
and not the directory where the binary executable files store or locate.

but you give me a valuable information that the user could test the bind10 in source tree, not under production directory,
then they will run "run_bind10.sh" instead of bind10 scripts.
So the bind10 in source tree should be also modified to solve the "same-out-file problem" if user runs bind 10 in source tree.

and thanks for your valuable suggestion. 


 





zhanghaikuo

From: JINMEI Tatuya /ÉñÃ÷ß_ÔÕ
Date: 2012-02-10 02:14
To: zhanghaikuo
CC: bind10-dev
Subject: Re: [bind10-dev] discussion about the approach of gprof
At Wed, 8 Feb 2012 10:18:28 +0800,
zhanghaikuo <zhanghaikuo at cnnic.cn> wrote:

> If the GMON_OUT_PREFIX is defined as you said,the rule of profiling
> data file name is "GMON_OUT_PREFIX"+"PID" (PID is process ID).When
> you run bind10,all the sub-processes will get the same value of
> GMON_OUT_PREFIX and you don't know sub-processes's PID, so you don't
> know which profiling data file is for b10-auth,and which is for
> b10-resolver.

Ah, okay, understood.

I have a different, but possibly related question/comment.  I don't
think it a common operation to install executables built with -pg due
to its severe performance impact.  At least it would be very unlikely
that they are installed under production directory.  I suspect a
common practice would be to run it within the source tree, using the
run_bind10.sh script.

I suggest noting this (and how to do it) in the wiki page.

Now, the possibly related question to the original point: do we still
have the same-output-file problem if run it within the source tree?
If not, we'll not have to use GMON_OUT_PREFIX, much less modifying the
bind10 script.  We might still mention the tweak to the bind10 script
in case someone really wants to run it under an installed environment,
but I guess it would be a supplemental note.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/bind10-dev/attachments/20120210/beac1860/attachment.html>