This is the mail archive of the
systemtap@sources.redhat.com
mailing list for the systemtap project.
RE: variables in scopes
- From: "Chen, Brad" <brad dot chen at intel dot com>
- To: "Richard J Moore" <richardj_moore at uk dot ibm dot com>, "Vara Prasad" <prasadav at us dot ibm dot com>
- Cc: "Ulrich Drepper" <drepper at redhat dot com>, "Frank Ch. Eigler" <fche at redhat dot com>, "Jim Keniston" <jkenisto at us dot ibm dot com>, "SystemTAP" <systemtap at sources dot redhat dot com>
- Date: Tue, 19 Apr 2005 09:34:42 -0700
- Subject: RE: variables in scopes
One conclusion I'd draw from Richard and Vara's comments
is that tapset authors would commonly reference kernel
data structures and so would want familiar C syntax, but
script authors would not, and in that context we might
sacrifice the convenience of C for something with better
safety properties. Vara, Richard, do you agreee/disagree?
Brad
-----Original Message-----
From: systemtap-owner@sources.redhat.com
[mailto:systemtap-owner@sources.redhat.com] On Behalf Of Richard J Moore
Sent: Tuesday, April 19, 2005 9:15 AM
To: Vara Prasad
Cc: Ulrich Drepper; Frank Ch. Eigler; Jim Keniston; SystemTAP
Subject: Re: variables in scopes
Vara Prasad wrote on 14/04/2005 14:57:32:
> I think if we go the route of full "C" syntax we will not be any where
> close to what Dtrace offers in the area of ease of use. The most value
i
> see in SystemTAP is knowledge of experts provided in the form of
> tapsets. The role of the language is an easy way to access those
tapsets
> and data they provide. Once we develop decent set of tapset libraries
> most people including kernel developers for most of the common tasks
> find what is available in the library sufficient. There might be
> occasions where one needs more than what is available and if you are a
> kernel developer you can always edit generated code and add what you
> need and make the module. If you are an end user even if the language
> allows without considerable kernel knowledge you wont be able to
enhance
> the library anyway so it doesn't matter.
>
> I would like to point out that recently in a customer situation some
of
> the best Linux Kernel experts who have developed entire filesystems
and
> major parts of the vm area etc. were trying to solve a performance
> problem along side the Sun folks. Sun folks using the Dtrace were able
> to narrow down the problem literally in less than 30 minutes and
Linux
> Kernel folks couldn't figure out even for days. These guys are
Kernel
> experts, their main feedback to me was your Kprobe stuff is cool but
not
> very useful without library that exports the data and a simple
scripting
> language. I think we should not forget that language role is to make
it
> easy to get what we want out of kernel. If we make this as a full
blown
> c language, then i see where little difference in writing systemtap
> scripts vs kprobe modules.
>
> Just my 2 cents.
>
> bye,
> Vara Prasad
>
Agreed. If one wants the write probes in C then why use an interpretive
form of C?
Surely one would write kernel modules that would call the kprobes KPIs
directly.