Feedback on my python framework I'm building.

[nbv...@gmail.com]

http://giotto.readthedocs.org/en/latest/tutorial.html

Can someone give me some feedback on what they think of this framework? I came up with the idea of this framework a few months ago. I gave a talk at a local python user group regarding these ideas, but no one seemed to think I was onto anything special or useful.

Basically its a framework that forces the developer(s) to strictly separate the model from the view and controller. You can 'hook up' multiple controllers to a project. The model layer can be completely mocked out so front end designers don't have to bother setting up Postgres/rabbitmq/whatever.

Does anyone have any throughts or feedback?

[Etienne Robillard]

Docs are useless for developers with an quarter of imagination. :-)

--
Etienne Robillard
Green Tea Hackers Club
Fine Software Carpentry For The Rest Of Us!
http://gthc.org/
er...@gthcfoundation.org

[Chris Angelico]

On Sat, Oct 13, 2012 at 3:49 PM, <nbv...@gmail.com> wrote:
> Basically its a framework that forces the developer(s) to strictly separate the model from the view and controller. You can 'hook up' multiple controllers to a project. The model layer can be completely mocked out so front end designers don't have to bother setting up Postgres/rabbitmq/whatever.

I don't like frameworks that force too much. Actually, I guess that
means I don't like frameworks at all, I like toolsets. Let the
programmer decide what he wants to do.

That said, though, there are times when a good framework can do 90% of
your work for you. The trouble comes when you want to do something the
author didn't think of - you might well end up either fighting against
the system, or modifying the framework to suit your task (and that
works only if you created it yourself). Thin frameworks are usually
immune to this, but on the flip side, they're less likely to be doing
most of your work for you.

It's really easy to demo something and show how awesome it is. How
easily can it be turned to a task it was never built for?

ChrisA

[Etienne Robillard]

On Sun, 14 Oct 2012 01:12:30 +1100
Chris Angelico <ros...@gmail.com> wrote:

> On Sat, Oct 13, 2012 at 3:49 PM, <nbv...@gmail.com> wrote:

> > Basically its a framework that forces the developer(s) to strictly separate the model from the view and controller. You can 'hook up' multiple controllers to a project. The model layer can be completely mocked out so front end designers don't have to bother setting up Postgres/rabbitmq/whatever.
>

> I don't like frameworks that force too much. Actually, I guess that
> means I don't like frameworks at all, I like toolsets. Let the
> programmer decide what he wants to do.
>
> That said, though, there are times when a good framework can do 90% of
> your work for you. The trouble comes when you want to do something the
> author didn't think of - you might well end up either fighting against
> the system, or modifying the framework to suit your task (and that
> works only if you created it yourself). Thin frameworks are usually
> immune to this, but on the flip side, they're less likely to be doing
> most of your work for you.
>
> It's really easy to demo something and show how awesome it is. How
> easily can it be turned to a task it was never built for?

Perhaps we deserve such lame/mediocre web frameworks after all, as Imagination is for those who
deserve the freedom to use it properly.

> ChrisA
> --
> http://mail.python.org/mailman/listinfo/python-list

[nbv...@gmail.com]

Do you have an example of a task that giotto can't handle that other frameworks can? One of my goals is to have this framework "turing complete" in the sense that everything that other frameworks can do, giotto should be able to do. I think my controller -> input_middleware -> model -> cache -> view -> output_middleware pattern pretty much allows for anything. Throughout the development process, I've radically changed the API many times when it was discovered that a certain task would be very hard (for instance authentication)

[nbv...@gmail.com]

On Saturday, October 13, 2012 10:13:22 AM UTC-4, Chris Angelico wrote:

[Etienne Robillard]

On Sat, 13 Oct 2012 08:57:47 -0700 (PDT)
nbv...@gmail.com wrote:


> Do you have an example of a task that giotto can't handle that other frameworks can? One of my goals is to have this framework "turing complete" in the sense that everything that other frameworks can do, giotto should be able to do. I think my controller -> input_middleware -> model -> cache -> view -> output_middleware pattern pretty much allows for anything. Throughout the development process, I've radically changed the API many times when it was discovered that a certain task would be very hard (for instance authentication)

> --
> http://mail.python.org/mailman/listinfo/python-list

how about eliminating poverty and employment discrimination?

Or can your script prevent childs to kill themselves on facebook?

let me know!

[Chris Angelico]

On Sun, Oct 14, 2012 at 2:57 AM, <nbv...@gmail.com> wrote:
> Do you have an example of a task that giotto can't handle that other frameworks can? One of my goals is to have this framework "turing complete" in the sense that everything that other frameworks can do, giotto should be able to do. I think my controller -> input_middleware -> model -> cache -> view -> output_middleware pattern pretty much allows for anything.

No, I don't, because I haven't tried to use it. But allow me to give
two examples, one on each side of the argument.

The 'tee' utility is primarily for writing a pipe to disk AND to
further pipelining, for instance:

./grind | tee grind.log | egrep '(2012|tps|@)'

It'll show me, in real-time, the lines that are of interest to me, but
it also writes the whole log to disk. But tee can be used in a quite
different way:

echo "Hello, world!" | sudo tee /some/file/owned/by/root

Unix is full of utilities of this nature. They each do one thing and
do it well, so when someone wants to use them in unusual ways, the
tools perform as expected.

On the contrary, Magic: The Gathering Online is designed for one
purpose: human interaction. There are a number of bot accounts that
function as shopkeepers, but their owners have to either use GUI level
monitoring and automation facilities, or hack the existing client
brutally. These people are trying to use an application in a way that
it wasn't designed to do, and no matter how reasonable the goal is,
the program just isn't built with toolchaining in mind.

> Throughout the development process, I've radically changed the API many times when it was discovered that a certain task would be very hard (for instance authentication)

This is, in my opinion, not a good thing. If you have to change your
API radically to support something you just thought of, then something
you still haven't thought of may require another radical API change.
That's not too bad if it's a personal system that you don't share with
anyone (though even then, you have to consider whether you'll be
spending more time fiddling with the framework than actually getting
stuff done), but is potentially disastrous in something you publish -
you can't afford to break everyone else's code, which means that
you'll be unable to support things you didn't think of.

The only way to support *absolutely everything* is to do nothing - to
be a framework so thin you're invisible. (That's not to say you're
useless; there are bridge modules that do exactly this - ctypes can
call on any library function from Python; it must therefore abstract
away nothing. But that's a very specific sort of case.) Anything else
must by nature make some things easier and some things harder.

The easiest way to avoid making things very much harder is to have a
way to ignore the framework - in the same way that most file
compression formats have a way to mark data as having not been
compressed (PKZip uses the term "Stored" rather than, eg, "Deflated").
This is why I say it's likely not a good thing that your framework
*forces* the separation of model/view/controller. You make it
impossible to temporarily ignore the framework. I haven't tested your
specific case, but what I've seen happen in similar situations is a
ridiculous set of "pass-through" methods or parameters, getting sent
up the line and back down the line, carrying opaque information that
really ought to have been manipulated at a different level in the
chain. I can't show you demo code as I don't own copyright on it, but
the git repository at work has, shall we say, a certain amount of
TheDailyWTF-level code in it...

But please don't take this post as _entirely_ doom and gloom. For a
number of reasons, I do not want to start coding within your
framework; but I would recommend that you DO (continue to) do so.
You'll learn a lot about what you do and don't like, and that's a good
thing. You'll quite possibly end up building something that saves you
a heap of time, and that's also a good thing. And at very worst, it'll
be an exploration. All programmers benefit from writing more code :)
But so long as you keep it personal, you're free to change it around
completely every time you need to. Keep that flexibility and you won't
feel constrained.

ChrisA

[nbv...@gmail.com]

On Saturday, October 13, 2012 12:48:23 PM UTC-4, Chris Angelico wrote:
> No, I don't, because I haven't tried to use it. But allow me to give
> two examples, one on each side of the argument.
>
> The 'tee' utility is primarily for writing a pipe to disk AND to
> further pipelining, for instance:

Could you please spent 10 minutes to read through the tutorial? The 'tee' unix utility and ctypes describes the way giotto goes about it business very well. Traditional web frameworks (such as django and rails) are too much of a 'magic the gathering' for my taste, which is why I'm writing giotto.

I'm really not looking for general "why I hate frameworks" criticism. I'm looking for specific criticism of the framework that I am writing.

> This is, in my opinion, not a good thing. If you have to change your
> API radically to support something you just thought of, then something
> you still haven't thought of may require another radical API change.

Not all api's get it right on the first shot. I'm more interested in getting it right, rather than patching together a bunch of random 'fixes' ala PHP.

> The only way to support *absolutely everything* is to do nothing - to
> be a framework so thin you're invisible. (That's not to say you're

You just described what Giotto is trying to do, since Giotto doesn't touch the model at all.

> This is why I say it's likely not a good thing that your framework
> *forces* the separation of model/view/controller. You make it
> impossible to temporarily ignore the framework.

Exactly. When you 'break out of the framework' you pile on technical debt. I want to force developers to not do that.

[nbv...@gmail.com]

On Saturday, October 13, 2012 12:48:23 PM UTC-4, Chris Angelico wrote:

> No, I don't, because I haven't tried to use it. But allow me to give
> two examples, one on each side of the argument.
>
> The 'tee' utility is primarily for writing a pipe to disk AND to
> further pipelining, for instance:

Could you please spent 10 minutes to read through the tutorial? The 'tee' unix utility and ctypes describes the way giotto goes about it business very well. Traditional web frameworks (such as django and rails) are too much of a 'magic the gathering' for my taste, which is why I'm writing giotto.

I'm really not looking for general "why I hate frameworks" criticism. I'm looking for specific criticism of the framework that I am writing.

> This is, in my opinion, not a good thing. If you have to change your
> API radically to support something you just thought of, then something
> you still haven't thought of may require another radical API change.

Not all api's get it right on the first shot. I'm more interested in getting it right, rather than patching together a bunch of random 'fixes' ala PHP.

> The only way to support *absolutely everything* is to do nothing - to
> be a framework so thin you're invisible. (That's not to say you're

You just described what Giotto is trying to do, since Giotto doesn't touch the model at all.

> This is why I say it's likely not a good thing that your framework
> *forces* the separation of model/view/controller. You make it
> impossible to temporarily ignore the framework.

[Chris Angelico]

On Sun, Oct 14, 2012 at 5:18 AM, <nbv...@gmail.com> wrote:
> On Saturday, October 13, 2012 12:48:23 PM UTC-4, Chris Angelico wrote:
>> No, I don't, because I haven't tried to use it. But allow me to give
>> two examples, one on each side of the argument.
>>
>> The 'tee' utility is primarily for writing a pipe to disk AND to
>> further pipelining, for instance:
>
> Could you please spent 10 minutes to read through the tutorial?

A fair criticism, and I am duly chastised. Okay. Now reading through
your web site... alright, I'm back.

>> This is why I say it's likely not a good thing that your framework
>> *forces* the separation of model/view/controller. You make it
>> impossible to temporarily ignore the framework.
>
> Exactly. When you 'break out of the framework' you pile on technical debt. I want to force developers to not do that.

Philosophy point 2: Giotto does force users to do things the “Giotto
way”. In other words, convention over configuration

Nice theory, but this is the bit that I fundamentally disagree with.
Forcing programmers to work in one particular style is usually not the
job of the language/framework/library. That should be up to the
programmer, or at least the local style guide. I do like your plan of
having identical interfaces to the same functionality (your example of
web-based and command-line is particularly appealing to my personal
tastes), but the same can usually be achieved with a well-built
library. In fact, all you need to do is have your model as a simple
Python module, and then the view and controller call on its functions.

What does your framework offer over a basic system like that?

ChrisA

[nbv...@gmail.com]

On Saturday, October 13, 2012 2:33:43 PM UTC-4, Chris Angelico wrote:
>
> Nice theory, but this is the bit that I fundamentally disagree with.
> Forcing programmers to work in one particular style is usually not the
> job of the language/framework/library. That should be up to the
> programmer, or at least the local style guide.

Have you ever read the zen of python? "Theres only one way to do it" is a core motto of the python language. In my opinion, this is the most important aspect of python and is what makes python so much better than PHP and perl and all the other "do it however you want, the more convoluted and obfuscated the better!" languages.

> I do like your plan of
> having identical interfaces to the same functionality (your example of
> web-based and command-line is particularly appealing to my personal
> tastes), but the same can usually be achieved with a well-built
> library. In fact, all you need to do is have your model as a simple
> Python module, and then the view and controller call on its functions.
>
> What does your framework offer over a basic system like that?

This "well built library" you mention basically describes my framework. You write a model method/function that takes data and then returns data. All giotto does is extract that data from the controller, pass it on to the model, then take the output of the model and pass it in to the view. You write the view, you write the model. Giotto provides the API for making al this happen. Giotto doesn't care if your model calls Postgres or Mysql or Redis or RabbitMQ, thats not of any concern to the framework.

The advantage of this is that the framework can 'mock' out the model layer very easily. For instance, your front end designers can work on the HTML without needing to run postgres, and the backend developers can work on the backend through the command line.

[nbv...@gmail.com]

On Saturday, October 13, 2012 2:33:43 PM UTC-4, Chris Angelico wrote:
>
> Nice theory, but this is the bit that I fundamentally disagree with.
> Forcing programmers to work in one particular style is usually not the
> job of the language/framework/library. That should be up to the
> programmer, or at least the local style guide.

Have you ever read the zen of python? "Theres only one way to do it" is a core motto of the python language. In my opinion, this is the most important aspect of python and is what makes python so much better than PHP and perl and all the other "do it however you want, the more convoluted and obfuscated the better!" languages.

> I do like your plan of
> having identical interfaces to the same functionality (your example of
> web-based and command-line is particularly appealing to my personal
> tastes), but the same can usually be achieved with a well-built
> library. In fact, all you need to do is have your model as a simple
> Python module, and then the view and controller call on its functions.
>
> What does your framework offer over a basic system like that?

[Chris Angelico]

On Sun, Oct 14, 2012 at 9:24 AM, <nbv...@gmail.com> wrote:
> On Saturday, October 13, 2012 2:33:43 PM UTC-4, Chris Angelico wrote:
>>
>> Nice theory, but this is the bit that I fundamentally disagree with.
>> Forcing programmers to work in one particular style is usually not the
>> job of the language/framework/library. That should be up to the
>> programmer, or at least the local style guide.
>
> Have you ever read the zen of python? "Theres only one way to do it" is a core motto of the python language. In my opinion, this is the most important aspect of python and is what makes python so much better than PHP and perl and all the other "do it however you want, the more convoluted and obfuscated the better!" languages.

Many times, but 'import this' doesn't translate into a language rule
that all classes have an uppercase first letter and all non-classes
don't; nor does it require that it be impossible to combine two simple
statements onto one line (because it's equally "obvious" to put them
onto two lines). Some things are questions of style, and are and
should be both implemented.

ChrisA

[Oscar Benjamin]

On 13 October 2012 17:48, Chris Angelico <ros...@gmail.com> wrote:
>
> The only way to support *absolutely everything* is to do nothing - to
> be a framework so thin you're invisible. (That's not to say you're

> useless; there are bridge modules that do exactly this - ctypes can
> call on any library function from Python; it must therefore abstract
> away nothing. But that's a very specific sort of case.) Anything else
> must by nature make some things easier and some things harder.

Without further ado I announce the release of my new library that
simultaneously does everything and nothing. It can be installed with:

pip install ''

That's right: the library's name is the empty string (as it's source
and documentation). Don't be fooled by the error message from pip: the
library was correctly installed before you ran the command!

Other libraries have failed by imposing restrictions on usage in an
attempt to better serve a mere finite number of use cases. This
library is *equally optimal* for an uncountably infinite number of
purposes and comes with the strong guarantee that it will never
decrease your productivity in any situation whatsoever!


Oscar

P.S. I also have not taken 10 minutes to read the documentation for
giotto as I get the impression that is not relevant to any of my own
use cases.

[Steven D'Aprano]

On Sat, 13 Oct 2012 15:24:04 -0700, nbvfour wrote:

> On Saturday, October 13, 2012 2:33:43 PM UTC-4, Chris Angelico wrote:
>>
>> Nice theory, but this is the bit that I fundamentally disagree with.
>> Forcing programmers to work in one particular style is usually not the
>> job of the language/framework/library. That should be up to the
>> programmer, or at least the local style guide.
>
> Have you ever read the zen of python? "Theres only one way to do it" is
> a core motto of the python language.

Have *you* ever read the Zen of Python? The line from the Zen is:

"There should be one-- and preferably only one --obvious way to do it."

Paraphrasing for emphasis:

There SHOULD be one (or more, but one is best) OBVIOUS way to do it.

as opposed to languages where there are no obvious ways to things, or
thirty.

Don't believe me that the emphasis is on *obvious* rather than "only"?
The very next line of the Zen tells you:

"Although that way may not be OBVIOUS at first unless you're Dutch."

[emphasis added]

Not being Dutch, I don't know whether the obvious way to do command line
argument handling is the getopt module or argparse. But there certainly
isn't *only one way* to do command line argument handling.

It is a gross canard, mostly spread by Perl programmers, that Python is
"limited" to "only one way to do it" and therefore isn't as good as Perl
which gives you "more freedom" (to write unreadable, unmaintainable code).

It simply isn't true that Python only gives you "only one way". The
acronym OOWTDI stands for *one obvious way to do it*.


--
Steven

[Tim Chase]

On 10/13/12 21:25, Steven D'Aprano wrote:
> Not being Dutch, I don't know whether the obvious way to do command line
> argument handling is the getopt module or argparse. But there certainly
> isn't *only one way* to do command line argument handling.

As an aside, I just watched a fascinating video[1] on docopt[2]
which seems like an amazingly simple way to build command-line UIs.
I haven't toyed with it yet, but I just wanted to share my "wow,
that's slick" experience with the list.

-tkc


[1]
http://www.youtube.com/watch?v=pXhcPJK5cMc

[2]
http://docopt.org/

[MRAB]

I think it's the "Paradox of Choice".

The more choices there are, the more time you'll spend trying to decide
which one is "best".

[Steven D'Aprano]

On Sun, 14 Oct 2012 05:33:40 +1100, Chris Angelico wrote:

> Forcing programmers to work in one particular style is usually not the
> job of the language/framework/library.

Have you actually programmed before?

*grin*

I've never come across a language/framework/library that DOESN'T force
programmers to work in a particular style, at least to some degree.

Every language encourages certain styles, discourages others, and makes
some impossible. Remember using PEEK and POKE commands with BASIC back in
1978? Pretty much impossible in Python. Perl probably has a way to do it
*wink* but few others do, and thank goodness we've moved on.

The sort of code you'll write in Fortran77 will be significantly
different from that you'll write in Lisp or Java or Forth or Ocaml.
Languages don't just differ in syntax, they differ in what they consider
good idioms, or even *possible* idioms. Hence:

http://dirtsimple.org/2004/12/python-is-not-java.html
http://dirtsimple.org/2004/12/java-is-not-python-either.html

Can you write "Java in Python"? Well, to some degree you can. Any Turing
complete language can be written in any other Turing complete language,
with sufficient layers of indirection, modulo things which are completely
impossible. You simply cannot do C-style pointer manipulations in Python,
not without breaking out of Python (e.g. with ctypes).

Libraries and frameworks are in a similar boat. Can you write CherryPy in
Django? Not easily. Generally the way to ignore a framework or library
and write in a different style is to, well, ignore the framework or
library and write in a different style :)

You can always step outside the bounds of the framework or library and
write something in the programming language level. Just because numpy
doesn't have some function I want, doesn't mean that I can't write it in
Python -- and just because numpy wants to work with arrays doesn't mean I
can't insist in using XML for my data, provided I convert it into an
array each time I pass it to a numpy function and back to XML when it
returns.

(For my next trick, watch as I eat three pounds of high-level radioactive
waste!)

Now clearly some frameworks are lighter than others, some impose their
style on your code like a big heavy weight on a rubber sheet, distorting
everything for miles around. Others not so much. But I really don't see
the problem here: if you don't like the "Twisted style" that it imposes,
don't use Twisted, use another framework, or no framework at all.

I think it is a *good* thing that different languages/frameworks/
libraries have their own styles. The alternative isn't "many styles" but
"one style": an unstylish mess.



--
Steven

[Chris Angelico]

On Sun, Oct 14, 2012 at 2:37 PM, Steven D'Aprano
<steve+comp....@pearwood.info> wrote:
> On Sun, 14 Oct 2012 05:33:40 +1100, Chris Angelico wrote:
>
>> Forcing programmers to work in one particular style is usually not the
>> job of the language/framework/library.
>
> Have you actually programmed before?
>
> *grin*
>
> I've never come across a language/framework/library that DOESN'T force
> programmers to work in a particular style, at least to some degree.

Style and technique aren't quite the same thing, though. Where you
draw the line between aspects the language should enforce and aspects
that should be up to the programmer is a tricky question. A language's
standard library certainly encourages a particular naming style, but
nothing enforces it.

Of course the language has to enforce something. I just don't think
that the MVC model should be part of what's enforced. What next?
Require that one file be one class and have nothing else in it?

ChrisA

[Roy Smith]

In article <507a3365$0$6574$c3e8da3$5496...@news.astraweb.com>,

Steven D'Aprano <steve+comp....@pearwood.info> wrote:

> Remember using PEEK and POKE commands with BASIC back in
> 1978? Pretty much impossible in Python.

But, trivial to implement as an extension :-)

[Dave Angel]

PEEK and POKE were intended to be used with memory mapped devices.
Simplest example is the 6502 chip, which had no I/O bus -- it was all
memory mapped. Want to change baud rate? poke a byte somewhere.

These days, the only device I can think of that's usually memory mapped
is the video. And few programs talk to it that way.

Now, INP and OUT (or various similar names) were for doing port I/o.
But I suspect that modern systems aren't going to let you do much of
that either.

--

DaveA

[MRAB]

On 2012-10-14 23:38, Dave Angel wrote:
> On 10/14/2012 08:48 AM, Roy Smith wrote:

> PEEK and POKE were intended to be used with memory mapped devices.
> Simplest example is the 6502 chip, which had no I/O bus -- it was all
> memory mapped. Want to change baud rate? poke a byte somewhere.
>
> These days, the only device I can think of that's usually memory mapped
> is the video. And few programs talk to it that way.
>
> Now, INP and OUT (or various similar names) were for doing port I/o.
> But I suspect that modern systems aren't going to let you do much of
> that either.
>

It depends on the CPU. Some have specialised instructions for I/O,
others don't.

[Roel Schroeven]

MRAB schreef:

> On 2012-10-14 23:38, Dave Angel wrote:
>> On 10/14/2012 08:48 AM, Roy Smith wrote:

>> PEEK and POKE were intended to be used with memory mapped devices.
>> Simplest example is the 6502 chip, which had no I/O bus -- it was all
>> memory mapped. Want to change baud rate? poke a byte somewhere.
>>
>> These days, the only device I can think of that's usually memory mapped
>> is the video. And few programs talk to it that way.
>>
>> Now, INP and OUT (or various similar names) were for doing port I/o.
>> But I suspect that modern systems aren't going to let you do much of
>> that either.
>>
> It depends on the CPU. Some have specialised instructions for I/O,
> others don't.

I think Roy Smith meant that the OS on modern systems generally prevents
direct access from user space to either memory mapped I/O or port mapped
I/O. Embedded systems might allow direct access, but AFAIK Python
doesn't run on those.

Best regards,
Roel

--
"Too often we hold fast to the cliches of our forebears. We subject all
facts to a prefabricated set of interpretations. Too often we enjoy the
comfort of opinion without the discomfort of thought."
-- John F Kennedy

ro...@roelschroeven.net