Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Richard Riehle]

"Thomas G. Marshall" <tgm2tothe10thpower@replacetextwithnumber.hotmail.com>
wrote in message news:3bOUc.46253$US4.14922@trndny01...
> Richard Riehle <adaworks@earthlink.net> coughed up the following:
> > Please give us an example of a type definition that
> > is difficult to catch and fix.
> >
> > Richard Riehle
>
> I don't understand what you're asking him.  Are you saying that you've
never
> seen a complicated model with a set of types that was ever so slightly
/off/
> the mark so that it was difficult to discover, and also laborious to fix?
>
There are several parts to your question.

    1) Complicated model
    2) Incorrect types ("ever so slightly /off/ the mark
    3) Difficult to discover
    4) Laborious to fix

My experienc iis primarily with Ada, so these questions are not quite
as relevant as they might be in some other language.  Nevertheless,
I will answer them.

I have seen designs where the there were too many types defined.  In
particular, people sometimes design too many floating point types.  For
example,

            package Real_Numbers is
                type Real is digits 8 range -2000.0 .. 2000.0;
                type Degree is digits 6 range 0.0 .. 360.0;
                -- and many more such definitions
           end Real_Numbers;

In the above example, it might be more useful to derived Degree from
Real, or create an Ada subtype (I will not define the difference here, but
it is different) from Real for Degree.   When there are too many variations
on floating point, one often has to do too much type conversion elsewhere
within the program.   However, Ada never lets you get this wrong.

As to incorrect types, this will most often occur in composite types,
especially
record types.  On might forget to include a component of the type in the
definition.  Since record types, in Ada, are most frequently defined as
"limited" types,  they are not part of the public part of a specification.
This makes it quite easy to correct them without disturbing the integrity
of the underlying specification.

As to being difficult to discover, I have not seen this very often in
programming
with Ada.  The language is designed so the compiler will quickly highlight
any
inconsistencies.   The compilation process is not based on textual
information
alone.   Each unit that depends on another unit requires the unit on which
it
depends is successfully compiled first.  The compiler will not even begin to
compile a unit unless its dependency relationships have been resolved.
This
lends itself to rapid discovery of problems with type definitions.

When we consider laborious to fix, I rarely find that to be a problem.  In
fact,
I cannot think of the last time (in nearly 20 years) where the problem with
a type was laborious to fix.   Certainly, when I was a novice there were
problems I could not easily resolve.   With experience, I have found that
well-designed Ada does not entertain me with the kind of mysterious
errors I sometimes encounter in other languages.

I realize that most readers do not benefit fromt the Ada static compilation
model.  Still, that model does have the appeal, to those who do enjoy
it, of making the type system a blessing rather than a nuisance.

Now.  I would still like to see some examples of type problems that
correspond
to the four points in your reply, above.

Richard Riehle

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Richard Riehle]

"Kelly Hall" <khall@acm.org> wrote in message
news:XtOUc.4483$QJ3.4254@newssvr21.news.prodigy.com...
> Jeff Brooks wrote:
>
> > The problem is static type checking is a very weak form of testing.
>
> That's because most programming languages have very weak type systems.
> I strongly advocate beefier type systems.
>
> As long as you're willing to write every test case you care about - no
> wonder you want to spend less time in the editor :)  As a formal methods
> guy, I'd much rather write one proof and ignore a whole class of test
> cases.  Ideally, my compiler would do most of the proofs for me.  But
> we're not there yet - more's the pity.
>
May I suggest you take a serious look at the Praxis Software SPARK
development model.  As "a formal methods guy," you will probably
appreciate the contribution of SPARK.

Richard Riehle

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Thomas G. Marshall]

Richard Riehle <adaworks@earthlink.net> coughed up the following:
> "Thomas G. Marshall"
> <tgm2tothe10thpower@replacetextwithnumber.hotmail.com> wrote in
> message news:3bOUc.46253$US4.14922@trndny01...
>> Richard Riehle <adaworks@earthlink.net> coughed up the following:
>>> Please give us an example of a type definition that
>>> is difficult to catch and fix.
>>>
>>> Richard Riehle
>>
>> I don't understand what you're asking him.  Are you saying that
>> you've never seen a complicated model with a set of types that was
>> ever so slightly /off/ the mark so that it was difficult to
>> discover, and also laborious to fix?
>>
> There are several parts to your question.
>
>     1) Complicated model
>     2) Incorrect types ("ever so slightly /off/ the mark
>     3) Difficult to discover
>     4) Laborious to fix
>
> My experienc iis primarily with Ada, so these questions are not quite
> as relevant as they might be in some other language.  Nevertheless,
> I will answer them.
>
> I have seen designs where the there were too many types defined.  In
> particular, people sometimes design too many floating point types.
> For example,
>
>             package Real_Numbers is
>                 type Real is digits 8 range -2000.0 .. 2000.0;
>                 type Degree is digits 6 range 0.0 .. 360.0;
>                 -- and many more such definitions
>            end Real_Numbers;
>
> In the above example, it might be more useful to derived Degree from
> Real, or create an Ada subtype (I will not define the difference
> here, but it is different) from Real for Degree.   When there are too
> many variations on floating point, one often has to do too much type
> conversion elsewhere within the program.   However, Ada never lets
> you get this wrong.
>
> As to incorrect types, this will most often occur in composite types,
> especially
> record types.  On might forget to include a component of the type in
> the definition.  Since record types, in Ada, are most frequently
> defined as "limited" types,  they are not part of the public part of
> a specification. This makes it quite easy to correct them without
> disturbing the integrity of the underlying specification.
>
> As to being difficult to discover, I have not seen this very often in
> programming
> with Ada.  The language is designed so the compiler will quickly
> highlight any
> inconsistencies.   The compilation process is not based on textual
> information
> alone.   Each unit that depends on another unit requires the unit on
> which it
> depends is successfully compiled first.  The compiler will not even
> begin to compile a unit unless its dependency relationships have been
> resolved.
> This
> lends itself to rapid discovery of problems with type definitions.
>
> When we consider laborious to fix, I rarely find that to be a
> problem.  In fact,
> I cannot think of the last time (in nearly 20 years) where the
> problem with a type was laborious to fix.   Certainly, when I was a
> novice there were problems I could not easily resolve.   With
> experience, I have found that well-designed Ada does not entertain me
> with the kind of mysterious
> errors I sometimes encounter in other languages.
>
> I realize that most readers do not benefit fromt the Ada static
> compilation model.  Still, that model does have the appeal, to those
> who do enjoy
> it, of making the type system a blessing rather than a nuisance.
>
> Now.  I would still like to see some examples of type problems that
> correspond
> to the four points in your reply, above.
>
> Richard Riehle

Without having to hand you an inch thick spec......ok.  Here:

Think java applet running on a pc.  That pc connected to a very complicated
scientific device, composed of a great many number producing things.  I'm
going to use the term "gui component" here, but I really mean more than
View,  but also the Model/Controller folded in as well.

One of my java clients needed a very complicated multi-layered gui, where
each of the visual components (meters, text fields, flashing lights, text
labels grayed out, red, or red/gray, or blinking), had to in real time
reflect what was happening to that component's counterpart within another
device.

A ton of meterable "things" in the device, and a ton of real time gui
elements in the java app reflecting the various values, states, warning
levels, what have you.  Many of the gui items further controlled other
"things" in the device, which caused only a cascade of changes in the gui
elements that were monitoring the components from the device.

For a huge part of the development, the gui components were designed around
their own threads.  One drawn LED that needed to be updated 5 times a
second, one thread.  A textfield next to it that needed to be updated 25
times a second, another thread.  Each of those threads ran in real time
collecting (and sometimes sending) to/from the device.

When the number of such continually active gui components threatened to
exceed several hundred, we shifted the model and from the top converted
every single component into an entry into something called the Dispatch.
Dispatch maintained a list(1) of lists(2) of components.  Each list(2) was a
list of components that needed updating a particular interval.  All the 50ms
updates went into one such list.  All slower 150ms updates went into
another, etc.  23 distinct interval types meant 23 entries in list(1), each
of which a list(2) of components.

Because it was done in Java, a statically typed language, there was a great
deal of energy spent in just turning the crank of making sure that the types
shifted from prior strategy to the current one.  I hope I've been clear
enough here.



-- 
Framsticks.  3D Artificial Life evolution.  You can see the creatures
that evolve and how they interact, hunt, swim, etc.
http://www.frams.alife.pl/

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Dmitry A. Kazakov]

On Thu, 19 Aug 2004 04:54:42 GMT, Thomas G. Marshall wrote:

> Without having to hand you an inch thick spec......ok.  Here:
> 
> Think java applet running on a pc.  That pc connected to a very complicated
> scientific device, composed of a great many number producing things.  I'm
> going to use the term "gui component" here, but I really mean more than
> View,  but also the Model/Controller folded in as well.
> 
> One of my java clients needed a very complicated multi-layered gui, where
> each of the visual components (meters, text fields, flashing lights, text
> labels grayed out, red, or red/gray, or blinking), had to in real time
> reflect what was happening to that component's counterpart within another
> device.
> 
> A ton of meterable "things" in the device, and a ton of real time gui
> elements in the java app reflecting the various values, states, warning
> levels, what have you.  Many of the gui items further controlled other
> "things" in the device, which caused only a cascade of changes in the gui
> elements that were monitoring the components from the device.
> 
> For a huge part of the development, the gui components were designed around
> their own threads.  One drawn LED that needed to be updated 5 times a
> second, one thread.  A textfield next to it that needed to be updated 25
> times a second, another thread.  Each of those threads ran in real time
> collecting (and sometimes sending) to/from the device.
> 
> When the number of such continually active gui components threatened to
> exceed several hundred, we shifted the model and from the top converted
> every single component into an entry into something called the Dispatch.
> Dispatch maintained a list(1) of lists(2) of components.  Each list(2) was a
> list of components that needed updating a particular interval.  All the 50ms
> updates went into one such list.  All slower 150ms updates went into
> another, etc.  23 distinct interval types meant 23 entries in list(1), each
> of which a list(2) of components.
> 
> Because it was done in Java, a statically typed language, there was a great
> deal of energy spent in just turning the crank of making sure that the types
> shifted from prior strategy to the current one.  I hope I've been clear
> enough here.

1. I would say that it was a design fault. The idea to use active objects
for GUI instruments is wrong. We used to design several similar projects
and never tried to go that way, because it is so obviously(?) wrong. GUI
instruments should be passive controlled by "referesh" threads from
outside. Also that they should not directly communicate with neither actors
nor sensors (devices), but do it through a middleware interface.

2. I don't see why re-design (an expensive thing) should be easier with
dynamic types. [ You have to change, say, 30% of the code. If that 30% is
largely about types, then dealing with Ada I would prefer that. Because the
compiler can help me there. If that changes are about implementations, then
it is usually more painful. Please, do not tell me that in Smalltalk that
will be 0%! You have a design fault. So it will be still same 30%
distributed all over the code. ]

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Jim Rogers]

"Thomas G. Marshall"
<tgm2tothe10thpower@replacetextwithnumber.hotmail.com> wrote in
news:moWUc.5028$de4.548@trndny07: 

> 
> Without having to hand you an inch thick spec......ok.  Here:
> 
> Think java applet running on a pc.  That pc connected to a very
> complicated scientific device, composed of a great many number
> producing things.  I'm going to use the term "gui component" here, but
> I really mean more than View,  but also the Model/Controller folded in
> as well. 
> 
> One of my java clients needed a very complicated multi-layered gui,
> where each of the visual components (meters, text fields, flashing
> lights, text labels grayed out, red, or red/gray, or blinking), had to
> in real time reflect what was happening to that component's
> counterpart within another device.
> 
> A ton of meterable "things" in the device, and a ton of real time gui
> elements in the java app reflecting the various values, states,
> warning levels, what have you.  Many of the gui items further
> controlled other "things" in the device, which caused only a cascade
> of changes in the gui elements that were monitoring the components
> from the device. 
> 
> For a huge part of the development, the gui components were designed
> around their own threads.  One drawn LED that needed to be updated 5
> times a second, one thread.  A textfield next to it that needed to be
> updated 25 times a second, another thread.  Each of those threads ran
> in real time collecting (and sometimes sending) to/from the device.
> 
> When the number of such continually active gui components threatened
> to exceed several hundred, we shifted the model and from the top
> converted every single component into an entry into something called
> the Dispatch. Dispatch maintained a list(1) of lists(2) of components.
>  Each list(2) was a list of components that needed updating a
> particular interval.  All the 50ms updates went into one such list. 
> All slower 150ms updates went into another, etc.  23 distinct interval
> types meant 23 entries in list(1), each of which a list(2) of
> components. 
> 
> Because it was done in Java, a statically typed language, there was a
> great deal of energy spent in just turning the crank of making sure
> that the types shifted from prior strategy to the current one.  I hope
> I've been clear enough here.

I think you may be painting with too broad of a brush. I see several
problems in your description. One problem was the choice of your initial
design. Applets typically run on a single processor. Unless you apply
very careful design, a multi-threaded application on a single processor
will usually run slower than a single-threaded application doing the
same conceptual work. Threading introduces overheads such as scheduling
and locking, which are not encountered in a single-threaded application.

Another problem is the choice of Java itself. Java threading is 
non-deterministic. The Java coding model mixes the concept of a type
and a thread to the point that they are very hard to separate if
threads are rejected late in the development process. Your task of
stripping out the threading would have been much easier if types and
threads were orthoginal to each other.

You still had some performance challenges in the new design, with
nested timing loops. It is possible that one of your list(2) lists
would get large enough that the processor/language combination could
not activate all events in a list(2) within the specified performance
boundary. You would get timing creep that could affect all the events
in your list of lists system. It is very important to balance nested
control loops. Once balanced, those control loops provide a difficult
challenge during maintenance. Any additions or deletions from any
list(2) could potentially upset the timing of all loops. Changes must
be made in delays for each nested loop level to compensate for the
changes in the size of lists.

Notice that these problems are independent of whether a language type
system is dynamic or static.

Jim Rogers

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Thomas G. Marshall]

Dmitry A. Kazakov <mailbox@dmitry-kazakov.de> coughed up the following:
> On Thu, 19 Aug 2004 04:54:42 GMT, Thomas G. Marshall wrote:
>
>> Without having to hand you an inch thick spec......ok.  Here:
>>
>> Think java applet running on a pc.  That pc connected to a very
>> complicated scientific device, composed of a great many number
>> producing things.  I'm going to use the term "gui component" here,
>> but I really mean more than View,  but also the Model/Controller
>> folded in as well.
>>
>> One of my java clients needed a very complicated multi-layered gui,
>> where each of the visual components (meters, text fields, flashing
>> lights, text labels grayed out, red, or red/gray, or blinking), had
>> to in real time reflect what was happening to that component's
>> counterpart within another device.
>>
>> A ton of meterable "things" in the device, and a ton of real time gui
>> elements in the java app reflecting the various values, states,
>> warning levels, what have you.  Many of the gui items further
>> controlled other "things" in the device, which caused only a cascade
>> of changes in the gui elements that were monitoring the components
>> from the device.
>>
>> For a huge part of the development, the gui components were designed
>> around their own threads.  One drawn LED that needed to be updated 5
>> times a second, one thread.  A textfield next to it that needed to
>> be updated 25 times a second, another thread.  Each of those threads
>> ran in real time collecting (and sometimes sending) to/from the
>> device.
>>
>> When the number of such continually active gui components threatened
>> to exceed several hundred, we shifted the model and from the top
>> converted every single component into an entry into something called
>> the Dispatch. Dispatch maintained a list(1) of lists(2) of
>> components.  Each list(2) was a list of components that needed
>> updating a particular interval.  All the 50ms updates went into one
>> such list.  All slower 150ms updates went into another, etc.  23
>> distinct interval types meant 23 entries in list(1), each of which a
>> list(2) of components.
>>
>> Because it was done in Java, a statically typed language, there was
>> a great deal of energy spent in just turning the crank of making
>> sure that the types shifted from prior strategy to the current one.
>> I hope I've been clear enough here.
>
> 1. I would say that it was a design fault. The idea to use active
> objects
> for GUI instruments is wrong. We used to design several similar
> projects
> and never tried to go that way, because it is so obviously(?) wrong.

Stay off your high horse.


> GUI instruments should be passive controlled by "referesh" threads
> from
> outside.

They were.  Actually, that's not "passive", that's /actively/ controlled and
controlling, because many of the gui elements were dual duty: they move as
accessors, but can be moved (by user) to mutate.


> Also that they should not directly communicate with neither
> actors nor sensors (devices), but do it through a middleware
> interface.

We did that too.  From the beginning.

When I'm refering to gui components, I'm refering to /all/ that it takes for
the gui component to work.  Remember, this has to be a quick sketch of what
was going on.

...[rip]...


-- 
http://www.allexperts.com is a nifty way to get an answer to just about
/anything/.

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Thomas G. Marshall]

Jim Rogers <jimmaureenrogers@worldnet.att.net> coughed up the following:
> "Thomas G. Marshall"
> <tgm2tothe10thpower@replacetextwithnumber.hotmail.com> wrote in
> news:moWUc.5028$de4.548@trndny07:
>
>>
>> Without having to hand you an inch thick spec......ok.  Here:
>>
>> Think java applet running on a pc.  That pc connected to a very
>> complicated scientific device, composed of a great many number
>> producing things.  I'm going to use the term "gui component" here,
>> but I really mean more than View,  but also the Model/Controller
>> folded in as well.
>>
>> One of my java clients needed a very complicated multi-layered gui,
>> where each of the visual components (meters, text fields, flashing
>> lights, text labels grayed out, red, or red/gray, or blinking), had
>> to in real time reflect what was happening to that component's
>> counterpart within another device.
>>
>> A ton of meterable "things" in the device, and a ton of real time gui
>> elements in the java app reflecting the various values, states,
>> warning levels, what have you.  Many of the gui items further
>> controlled other "things" in the device, which caused only a cascade
>> of changes in the gui elements that were monitoring the components
>> from the device.
>>
>> For a huge part of the development, the gui components were designed
>> around their own threads.  One drawn LED that needed to be updated 5
>> times a second, one thread.  A textfield next to it that needed to be
>> updated 25 times a second, another thread.  Each of those threads ran
>> in real time collecting (and sometimes sending) to/from the device.
>>
>> When the number of such continually active gui components threatened
>> to exceed several hundred, we shifted the model and from the top
>> converted every single component into an entry into something called
>> the Dispatch. Dispatch maintained a list(1) of lists(2) of
>>  components. Each list(2) was a list of components that needed
>> updating a particular interval.  All the 50ms updates went into one
>> such list. All slower 150ms updates went into another, etc.  23
>> distinct interval types meant 23 entries in list(1), each of which a
>> list(2) of components.
>>
>> Because it was done in Java, a statically typed language, there was a
>> great deal of energy spent in just turning the crank of making sure
>> that the types shifted from prior strategy to the current one.  I
>> hope I've been clear enough here.
>
> I think you may be painting with too broad of a brush. I see several
> problems in your description. One problem was the choice of your
> initial design. Applets typically run on a single processor. Unless
> you apply very careful design, a multi-threaded application on a
> single processor will usually run slower than a single-threaded
> application doing the same conceptual work. Threading introduces
> overheads such as scheduling and locking, which are not encountered
> in a single-threaded application.
>
> Another problem is the choice of Java itself. Java threading is
> non-deterministic. The Java coding model mixes the concept of a type
> and a thread to the point that they are very hard to separate if
> threads are rejected late in the development process. Your task of
> stripping out the threading would have been much easier if types and
> threads were orthoginal to each other.
>
> You still had some performance challenges in the new design, with
> nested timing loops. It is possible that one of your list(2) lists
> would get large enough that the processor/language combination could
> not activate all events in a list(2) within the specified performance
> boundary. You would get timing creep that could affect all the events
> in your list of lists system. It is very important to balance nested
> control loops. Once balanced, those control loops provide a difficult
> challenge during maintenance. Any additions or deletions from any
> list(2) could potentially upset the timing of all loops. Changes must
> be made in delays for each nested loop level to compensate for the
> changes in the size of lists.
>
> Notice that these problems are independent of whether a language type
> system is dynamic or static.

{sigh}.  Design 101 was not lost on us.  Neither is Common Sense 801.  I
don't have the energy to discuss the design in its entirety, nor walk
through all the decisions, the reasoning, the background, the customer
demands, etc., etc.----it'd be unreasonable to assume that such a
conversation would even be /possible/ over usenet.

I was asked for an example of:

<quotezilla>
    1) Complicated model
    2) Incorrect types ("ever so slightly off the mark
    3) Difficult to discover
    4) Laborious to fix
</quotezilla>

The reasons /why/ these decisions were made?  The discussion of it per se,
is of zero importance.  Please, save your time and mine---you'll /never/ get
the full picture---this was a large project with many layers of design.





-- 
http://www.allexperts.com is a nifty way to get an answer to just about
/anything/.

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Stephen Leake]

"Thomas G. Marshall" <tgm2tothe10thpower@replacetextwithnumber.hotmail.com> writes:

> <snip description of complex GUI / instrument system>
> 
> Because it was done in Java, a statically typed language, there was a great
> deal of energy spent in just turning the crank of making sure that the types
> shifted from prior strategy to the current one.  I hope I've been clear
> enough here.

I gather you believe it would have been easier to do this redesign in
some other language? What language, and why?

If I had done the same thing in Ada (and I have done similar redesigns
(I call it "refactoring")), I would expect to "turn the crank" to get
all the types right. That's part of what static typing is for; the
compiler lets you know where stuff has to be changed. But at each
point, you need to make sure that nothing _else_ needs to be changed;
often it does.

So I don't see why you feel "turning the crank" is a Bad Thing here.

-- 
-- Stephe

Re: Static vs. Dynamic typing (big advantage or not)---WAS: c.programming: OOP and memory management

[Richard Riehle]

"Thomas G. Marshall" <tgm2tothe10thpower@replacetextwithnumber.hotmail.com>
wrote in message news:moWUc.5028$de4.548@trndny07...
> Richard Riehle <adaworks@earthlink.net> coughed up the following:
> > "Thomas G. Marshall"
>
> When the number of such continually active gui components threatened to
> exceed several hundred, we shifted the model and from the top converted
> every single component into an entry into something called the Dispatch.
> Dispatch maintained a list(1) of lists(2) of components.  Each list(2) was
a
> list of components that needed updating a particular interval.  All the
50ms
> updates went into one such list.  All slower 150ms updates went into
> another, etc.  23 distinct interval types meant 23 entries in list(1),
each
> of which a list(2) of components.
>
> Because it was done in Java, a statically typed language, there was a
great
> deal of energy spent in just turning the crank of making sure that the
types
> shifted from prior strategy to the current one.  I hope I've been clear
> enough here.
>
Yes, you are clear.  I see that, with Java, and the approach to design
you describe, you could get annoyed with the static typing, the length
of time to change things, and the difficulty of finding the exact problem
in each case.    I also see that, with Ada, designing the types in well
formed packages and separating the visible part of a type from its
structure (i.e., the name of the type from how it is defined), this
problem is largely non-existent.

Thanks.

Richard Riehle