Call for ideas for future musl-related talks

Call for ideas for future musl-related talks

[Rich Felker]

After having done a couple conference talks already at Ohio LinuxFest
2013 and 2014, I'm considering pursuing more conferences, but I'm not
sure what topics/framing would be most interesting and effective at
getting more people interested in musl. If there's anything special
you'd like to hear me give a talk on, or think would be constructive
to the project, reply and let me know.

Rich

Re: Call for ideas for future musl-related talks

[Justin Cormack]

On Mon, Dec 15, 2014 at 5:04 AM, Rich Felker <dalias@libc.org> wrote:
> After having done a couple conference talks already at Ohio LinuxFest
> 2013 and 2014, I'm considering pursuing more conferences, but I'm not
> sure what topics/framing would be most interesting and effective at
> getting more people interested in musl. If there's anything special
> you'd like to hear me give a talk on, or think would be constructive
> to the project, reply and let me know.

Apologies for not getting back sooner.

I think the most interesting topic for a talk for a generalist
audience is to cover the kinds of bugs you write about on ewontfix. (I
wouldn't talk about systemd though, it is too partisan for people to
listen clearly).

The focus should be around techniques for writing better software,
better specifications, and how to find problematic areas. And about
how writing tests for these things is hard, because a lot of them are
races, although talking about where tests do and dont work is good
too.

A title could be:
Finding bugs in glibc by writing a better libc
Better code by design, thought, and hard work

A structure like:
1. What is Musl and why did I start writing it
2. Libc bugs are like compiler bugs they really ruin your day
3. Go into some detail as per ewontfix on 1-3 bugs as per ewontfix
depending on talk length
4. How to spot potential bugs: code inspection, tests
5. Why Musl is less buggy than glibc: size, consistency, structure etc
6. Working with specifications, feedback, clarification
7. Musl is great, how to get started with it

Happy to feedback on any drafts.

Justin

Re: Call for ideas for future musl-related talks

[Szabolcs Nagy]

* Justin Cormack <justin@specialbusservice.com> [2014-12-28 16:24:43 +0000]:
> On Mon, Dec 15, 2014 at 5:04 AM, Rich Felker <dalias@libc.org> wrote:
> > After having done a couple conference talks already at Ohio LinuxFest
> > 2013 and 2014, I'm considering pursuing more conferences, but I'm not
> > sure what topics/framing would be most interesting and effective at
> > getting more people interested in musl. If there's anything special
> > you'd like to hear me give a talk on, or think would be constructive
> > to the project, reply and let me know.
> 
> Apologies for not getting back sooner.
> 
> I think the most interesting topic for a talk for a generalist
> audience is to cover the kinds of bugs you write about on ewontfix. (I
> wouldn't talk about systemd though, it is too partisan for people to
> listen clearly).
> 
> The focus should be around techniques for writing better software,
> better specifications, and how to find problematic areas. And about
> how writing tests for these things is hard, because a lot of them are
> races, although talking about where tests do and dont work is good
> too.
> 

my whishlist is

- why do posix and c matter in the age of web/mobile/cloud

- good/bad/ugly parts of posix/linux/toolchain from libc pov

- metrics (benchmarks, size, complexity, amount of libc code
executed in various use-cases, time spent in libc, etc)

Re: Call for ideas for future musl-related talks

[Rich Felker]

On Mon, Dec 29, 2014 at 01:29:30AM +0100, Szabolcs Nagy wrote:
> my whishlist is
> 
> - why do posix and c matter in the age of web/mobile/cloud
> 
> - good/bad/ugly parts of posix/linux/toolchain from libc pov
> 
> - metrics (benchmarks, size, complexity, amount of libc code
> executed in various use-cases, time spent in libc, etc)

I like your ideas, and IMO this is potentially interesting as a
less-technical talk -- or at least one that's compelling to a less
technical audience but that still gives an opportunity to bring in
technical material.

The "why does it matter" part is nice for bringing up the pitfalls of
container deployment with glibc and mainstream library stacks --
circular dependencies and difficulty of reproducible builds, need for
large supporting infrastructure in your containers, large bug/attack
surface, etc. Of course this doesn't really cover the matter of why
C/POSIX are needed (versus e.g. a core OS built on node.js or lua or
similar) but that's an opportunity to talk about resource accounting
and how hard it is to make a robust system groundwork in a language
that doesn't have a concept of storage and storage duration.

Rich

Re: Call for ideas for future musl-related talks

[Laurent Bercot]

On 30/12/2014 20:45, Rich Felker wrote:

> The "why does it matter" part is nice for bringing up the pitfalls of
> container deployment with glibc and mainstream library stacks --
> circular dependencies and difficulty of reproducible builds, need for
> large supporting infrastructure in your containers, large bug/attack
> surface, etc. Of course this doesn't really cover the matter of why
> C/POSIX are needed (versus e.g. a core OS built on node.js or lua or
> similar) but that's an opportunity to talk about resource accounting
> and how hard it is to make a robust system groundwork in a language
> that doesn't have a concept of storage and storage duration.

  If you do that talk, I would love to have a video of it, or at least
the slides. I often get asked this question about the skarnet.org
project and it's really hard to convince people of the importance of
smallness and reduced dependencies in the age of powerful computers
and packaged distribution; and the mere concept of attack surface isn't
even well understood. Having a document of reference to point people to
would be a tremendous help - and I've never taken the time to write a
serious one.

  If at some point you're open to external participation for the
preparation of such a talk or document, I'm very much willing to help.

-- 
  Laurent

Re: Call for ideas for future musl-related talks

[Natanael Copa]

On Mon, 29 Dec 2014 01:29:30 +0100
Szabolcs Nagy <nsz@port70.net> wrote:

> * Justin Cormack <justin@specialbusservice.com> [2014-12-28 16:24:43 +0000]:
> > On Mon, Dec 15, 2014 at 5:04 AM, Rich Felker <dalias@libc.org> wrote:

> > > If there's anything special you'd like to hear me give a talk on,
> > > or think would be constructive to the project, reply and let me
> > > know.
...
> my whishlist is
> 
> - why do posix and c matter in the age of web/mobile/cloud

+1

This could maybe include a few words on why standards in general matter
at all. (for example even the big company that gave us Internet
Explorer 6 realizes that it hurts themselves to not follow html
standard in the long run.)

I think this is a topic that people need to hear. Why standards are
good and why its a bad idea to let one implementation rule them all and
be the "standard" that all other implementations should mimic.

Imagine if all browsers should follow the IE 6 implementation of
html/css. At that time it maybe looked like a good idea to use all the
extra features to have your product compete with those who aimed for
"lowest common denominator" feature set - but today most people that
depends on a browser realize that was a bad idea in the long run.

> - good/bad/ugly parts of posix/linux/toolchain from libc pov
> 
> - metrics (benchmarks, size, complexity, amount of libc code
> executed in various use-cases, time spent in libc, etc)

Re: Call for ideas for future musl-related talks

[Rich Felker]

On Sun, Dec 28, 2014 at 04:24:43PM +0000, Justin Cormack wrote:
> On Mon, Dec 15, 2014 at 5:04 AM, Rich Felker <dalias@libc.org> wrote:
> > After having done a couple conference talks already at Ohio LinuxFest
> > 2013 and 2014, I'm considering pursuing more conferences, but I'm not
> > sure what topics/framing would be most interesting and effective at
> > getting more people interested in musl. If there's anything special
> > you'd like to hear me give a talk on, or think would be constructive
> > to the project, reply and let me know.
> 
> Apologies for not getting back sooner.

No problem. Thanks for the feedback/ideas!

> I think the most interesting topic for a talk for a generalist
> audience is to cover the kinds of bugs you write about on ewontfix. (I
> wouldn't talk about systemd though, it is too partisan for people to
> listen clearly).

I agree completely and I've omitted systemd from past talks except
possibly some brief mention in response to questions from the audience
(I forget whether my remarks on systemd were during or after the
sessions but they weren't inflammatory anyway :).

> The focus should be around techniques for writing better software,
> better specifications, and how to find problematic areas. And about
> how writing tests for these things is hard, because a lot of them are
> races, although talking about where tests do and dont work is good
> too.

This has a lot of overlap with my Ohio LinuxFest 2013 talk, and while
I think people found it interesting, I think it was too
developer-oriented for most of the audience to get a lot out of it. If
I do that type of talk again I'd want either to make sure I'm talking
to an audience with the appropriate technical background or to find a
way to make it less technical but still compelling (and IMO that's
really hard).

> A title could be:
> Finding bugs in glibc by writing a better libc
> Better code by design, thought, and hard work
> 
> A structure like:
> 1. What is Musl and why did I start writing it
> 2. Libc bugs are like compiler bugs they really ruin your day
> 3. Go into some detail as per ewontfix on 1-3 bugs as per ewontfix
> depending on talk length
> 4. How to spot potential bugs: code inspection, tests
> 5. Why Musl is less buggy than glibc: size, consistency, structure etc
> 6. Working with specifications, feedback, clarification
> 7. Musl is great, how to get started with it

I like this structure.

Rich

Re: Call for ideas for future musl-related talks

[Rich Felker]

On Mon, Dec 15, 2014 at 12:04:32AM -0500, Rich Felker wrote:
> After having done a couple conference talks already at Ohio LinuxFest
> 2013 and 2014, I'm considering pursuing more conferences, but I'm not
> sure what topics/framing would be most interesting and effective at
> getting more people interested in musl. If there's anything special
> you'd like to hear me give a talk on, or think would be constructive
> to the project, reply and let me know.

I wasn't aware until today, but Embedded Linux Conference 2015 will be
taking place this March in San Jose, CA and the deadline for proposals
is next week. I'm not sure if I'll be able to come up with a proposal
for it or get funding for travel, but even if I don't I think this
might be a nice opportunity for others with musl-based embedded
projects. Details are at:

http://free-electrons.com/blog/elc-2015-cfp/
http://events.linuxfoundation.org/events/embedded-linux-conference/program/cfp

Rich