[NEED HELP] Evangelise async design

[hypatia]

This is one of the suggestions from issue 1015.

"Answer the question; 'What's the point of going to all this trouble for an
async design when 'regular' coding is so much easier?' Even a simple link to
the C10k page would help ('Why Use Twisted' is a good spot for it).  Lots of
people hear that Twisted is a great networking toolkit, but come expecting
something more akin to Net::X modules on CPAN and not a plunge into learning
about asynchronous design."

This is actually not an area I have expertise in. There is some scattered
justification in the docs that seems to centre around race conditions, deadlocks
and the GIL. I'll see what I can come up with in unifying this. Suggestions
appreciated.

[itamarst]

1. scalability; see c10k page, plus python maks it worse since threads don't run
in parallel ("GIL") so don't get benefits from multiple CPUs.
2. lack of race conditions.
3. Some protocols are really hard to do with blocking APIs (e.g. think about how
you'd implement chat server if you had only threads and blocking read and write).

[itamarst]

Or well, less race conditions. Makes concurrency a much easier problem since you
can;t be interrupted at any point, only at event handler borders (or in defgen
when yielding Deferred).

[hypatia]

Thanks Itamar, that should be enough for a respectable write-up.

[itamarst]

I hope that wasn't sarcasm? :) I was going to add more detail later if needed.

[hypatia]

Not sarcasm! The more info I have the easier it is, but I should have the minium
required now to write something sensible.

[kragen]

Is there a URL where the result of this effort will appear?

The best thing I've found so far along these lines is Zooko's 2001 article, "Why
Multithreading Sucks, Using Mojo Nation As An Example," at
http://www.eros-os.org/pipermail/e-lang/2001-July/005410.html

[Glyph]

Not really a problem with Twisted as such, more an issue with programmers' brains.

This should just be done as necessary in the various documents that describe how the reactor works. I don't see how we could ever close this ticket.