Re: [boost] [Boost-docs] Maintaining boost.org

Subject: Re: [boost] [Boost-docs] Maintaining boost.org
From: OvermindDL1 (overminddl1_at_[hidden])
Date: 2010-06-02 15:00:21

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> otherwise
>> you are
>> free to kill the session once served.  The amount of
>> data it holds is
>> only your own design, can be as little or large as
>> necessary.
>
> There are two points:
>
> 1. Without caching (and AFAIK Wt does not even has such concept)
> Â  you'll get stuck very fast. Even C++ is faster then PHP by
> Â  order of magnitude this will not save you with large DB when
> Â  queries are not so cheep.

Wt is designed to be outright cached by an external caching program,
but it is trivial (and I have done so) to embed caching yourself (and
by trivial, I mean it took less then 5 minutes to code up).

Quite true with the DB queries, but I am not using C++ here for its
speed, I am using it because it is powerful, I know it, and it can
potentially do anything that I could ever imagine needing to do
anywhere, to get around big DB queries is a design issue, not a
language issue.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> 2. Web is stateless for 99% percent of visitors. Giving a session
> Â  to each request IMHO is bad idea.

Exactly, there is no point, designing most of it to be stateless would
be fine, although even holding state would not matter much, even
assuming some *massive* per-user session sizes of 200kb each still
lets it handle far more people then the Boost site would ever get at
any single instant.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> True, but it can handle a very Qt style or design (which I
>> personally
>> do not like, nor do I use it), or a very CSS centric style,
>> or any
>> mix.
>
> As I told you, IMHO web is not GUI... Thinking of it as GUI just
> makes things bad.

I beg to differ, what is HTML if not a way of rendering an interface,
and CSS to design a theme, and javascript to make it interactive?  Web
in general is not a GUI, just as the computer in general is not a GUI,
it is just the primary (not the only) way of interacting with the
system.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> Oh not even, the main advantage of web design in C++ is its
>> powerful
>> capabilities and ability to abstract things away.
>
> This is not about C++, it is about framework.

Exactly, and Wt can do just about anything you need, plus if you are
so against C++, use Wt in Java, it is exposed to other languages, not
just C++, C++ just happens to be what it is made in.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> Since I started
>> using Wt I can get a site running faster then I ever could
>> have with
>> PHP or Python
>
> I'm not talking about plain PHP, or plain Python. I'm talking about
> using web framework like Django.

As mentioned below, DJango and others are fine, but if you ever really
have used them, you find that you run into limitations very quickly,
in addition to creating something in Wt is faster due to the need to
create templates in those others.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> What I meant by low level is that it has no higher level
>> abstractions
>> as Wt has, it is nothing but pure MVC, where Wt is more
>> like Qt (which
>> is MVS anyway, but like a GUI program, or you can use the
>> lower-level
>> capabilities if you so wished).  CppCMS does *not* do
>> near what Wt
>> has, I tried it (I do prefer its license over Wt's),
>
> The point that Wt tries to make web development as GUI.

Only for the parts of it that actually are, if you need to make RSS
feeds, serve some static random content, generate dynamic images, or
anything else you can imagine, it is made for that as well.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> I think that web developer should write HTML, should write JavaScript
> (actually using one of JavaScript toolkits like JQuery) and
> write server side code.

Why should they write HTML?  Would you prefer to use the Win32 C API
for creating graphical programs instead of wxWidgets, MFC, Forms, Qt,
etc...?  I certainly would not, I want to get things done fast and
have it work everywhere.  If I wrote the HTML myself then I would have
to deal with all the little bugs from all the different browsers, when
Wt handles that all for me.  Besides, I do not have to use its QT-like
interface, it is trivial to send HTML too, but then you lose
advantages like only changing the parts of the page that need updating
instead of refreshing everything.  Oh, and yes, Wt does use jQuery
internally, and it makes it trivial to connect Javascript and C++
together if you have any custom code that needs to be run client-side
(as I had a lot of in one of my projects).

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> Hiding this behind the "GUI" framework leads to bad design.

So wxWidgets, Forms, Qt, GTK, etc... all encourage bad design?  I just
do not see that, where is the proof?

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> HTTP connections are not cheap, the signals/slots design make designer
> to forget when comes behind the scenes.

Exactly, they are not cheap, Wt minimizes the need of them.  If you
just send a static page, it compresses it in any necessary way.  If
you send dynamic pages, and the user does a lot of updating in the
page that requires lots of little things updated, then Wt combines all
the request into one large response and sends in en bulk (compressed
of course).  Using the signals/slots design lets it combine requests
and reduce them very well, far better then any plain HTML setup.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> Signals and slots are good when they called withing same process and
> not over RPC that is passed over HTTP.

Pure signals/slots in Wt are server-side only, designed to be called
within the same process, even by different users interacting with each
other (see the chat example at
http://www.webtoolkit.eu/wt/examples/simplechat/simplechat.wt ), there
are two types of external 'callbacks', both which are transparently
converted to use server-side signals, and two types of external sent
'signals'.  The two callbacks are when a user clicks a url link or a
form submit, Wt handles converting that to proper callbacks on the
server-side if necessary, and the other is a javascript link, which is
the basic type of Javascript callback, such as when you need to do
something that cannot be done in javascript (easily) and you want to
do it server-side instead, you just create a Javascript slot (jslot)
and register a C++ callback method (or more) with it and in your
output Javascript code on the webpage you just do this:
  ...some javascript code and so forth;
  " << myjslot << "(any arguments);
  ... some more javascript code;
and when that javascript function is called in that position then it
causes an RPC callback into the server (combined with any other
pending requests of course) to do what it wants, which may or may not
cause a client-side change if necessary.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> This abstraction leads to bad code, bad design and total misunderstanding
> what web is.

I do not see how, where is the proof?  Just as the old antiquated
windowing systems that are still usable on modern computers exist, no
one really uses them, there are always abstractions on top that allow
for things like vastly enhanced code readability, easy to add in
features, easy updating without changing massive amounts of code-base,
etc...

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> Good caching is useful, no matter what you use.
>>
>
> AFAIK Wt does not even has such concept... While it is one of the
> central design concepts of CppCMS - efficient caching.

It does not have one built-in, it is designed to be used with external
caching systems, but it is still extremely easy to create your own
inside of it, if you really want Wt to come with one then I can submit
mine as a patch once I clean it up a bit, nothing special.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>>Â One major difference is that you do not need
>> to write
>> templates as in CppCMS (well, you could also write out raw
>> html from
>> CppCMS), since Wt lets you design it like a Gui app, you
>> can create
>> things *FAR* quicker
>
> I disagree with you that writing HTML slows things down. It makes
> them clearer, easier to debug and smaller.

And I completely disagree with that.  Writing HTML compared to writing
*nothing* could in no way be faster.  It does not make things clearer,
it only adds in more noise with all the other css and javascript crap
in the way, and it in fact makes it vastly *harder* to debug due to
the way different browsers are broken in different ways, better to not
write it and let something else that already knows how to handle all
the broken stuff automatically.  And again, writing HTML compared to
writing *nothing*, noticeable size different there for sure.  Also, do
note, Wt has a template engine called WTemplate, although you could
use any other, or do not use it at all, I have never used it, it does
support displaying arbitrary widgets of any complexity too, so you
still have the full power of Wt with whatever you like about
templates.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> So on and so forth.  I tried using
>> CppCMS, but it just
>> does not do anywhere near as much to smooth the workflow
>> in comparison.
>
> - When had you tried it?
> - Which version?
> - What had you written with it?
>
> And:
>
> - How much time would it take to write an application like Wikipp in Wt?
> - What performance will you get?

I tried it when you introduced it to boost when talking about
Boost.Locale a few months back, unsure what the version was.  I just
followed the tutorials then tried to write something interactive as a
test (a copy of something I made in Wt a while back).
How much time to write an application like Wikipp in Wt, hmm, very
short no doubt, I can do so if you wish, but let me look at the Wikipp
code first, done, first I see no way of compiling, no cmake build
scripts or bjam or anything that will work on the computer I am on
now, so just looking at the source.

First of all, I notice a templates and mo/p4 directories, I do admit
that I have become a bit allergic to templates as of late due to the
ease of use of not needing them now (although you can still do them in
Wt no doubt), but I do notice that whatever template language that is,
it is quite hard to read, all this template code and HTML mixed
together, it seems you have to manually handle sidebars and so forth,
easy potential for things to break, and the generated cpp from it is
even harder to read, while no doubt requiring a pre-processing step.
As for the mo/p4 directories, obviously for internationalization.
Wt's, I admit, is not as advanced as Boost.Locale (although they will
probably use Boost.Locale should it end up in Boost anyway), but
theirs is made up of simple xml files, different directories for
different languages, different files for different sections of the
app, not just for localization, but can also be used for theming in a
different way letting you use arbitrary html and so forth (if you let
it).  It would be simple to use Boost.Locale within Wt as-is though,
it does not absolutely mandate styles.

As for the actual code, the main application seems to be about as long
as the Wt version would, although I wonder how this line works:
        url.add("^/(\\w+)(/.*)$",
                boost::bind(&wiki::run,this,$1,$2));

I notice there are a lot of things that are done in CppCMS that take a
lot more code to do, like a simple redirect takes two lines (which
includes instancing a HTTPRedirectHeader), compared to a simple
app->redirect in Wt (which can take a url or widget location, where
CppCMS seems to only take a url).

Here is the important part of the page display function:
"""
void page::display(string slug)
{
        this->slug=slug;

        sql<<	"SELECT title,content,sidebar FROM pages WHERE lang=? AND slug=?",
                locale,slug;
        row r;
        if(!sql.single(r)) {
                string redirect=edit_url();
                set_header(new HTTPRedirectHeader(redirect));
                add_header("Status: 302 Found");
                return;
        }
        ini(c);
        r >> c.title >> c.content >> c.sidebar;
        render("page",c);
}
"""

This same function in Wt might be this (using its own DB system, which
is still rather new, but you could use any other if you wanted):
"""
void page::display(string slug)
{
        this->slug=slug;

        dbo::Transaction t(*session_);
        dbo::ptr<Page> page = session.find<Page>().where("lang = ? AND slug=?")
                .bind(locale).bind(slug);

        if(!page) {
                string redirect=edit_url();
                app->redirect(redirect);
                return;
        }
        title->setText(page.title); // This is set to show html as text
        content->setText(page.content); // This is set to pass through html,
after making sure it is clean.
}
"""

The rest of the functions also look like they would go quite fast to convert.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> It took me about several evenings only to write such application from
> the begging to the end including testing, and deploying and fixing CppCMS
> bugs I had found.
>
> And it was first serious CppCMS application.

Making such a Wiki would probably take even less time, and add an RSS
feed for latest updates in one more function, and etc...

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> It holds about 2,500 requests per second on 1CPU machine.

I can see the Wiki being very DB constrained, so it would be tough to
compare this, perhaps a dynamic non-DB page generation request?
Do note, Wt's DB is very low level, using constructed queries and
other such optimizations, all available for re-use without needing to
regenerate.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
>> Ditto, and a couple years of experience with Wt, and a
>> decade with
>> PHP, and a **few months** with CppCMS.
>
> Good point.

I was searching for C++ web frameworks a while ago, CppCMS never came
up, so that is why I did not notice it until your Boost post.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> Summary:
> --------
>
> - Developing web application as you develop GUI is not suited for
> Â general web development

Oh it very much is well suited for that, well, if you are making an
interface, for RSS feeds, image generation, *whatever*, Wt has things
for that too.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> - Signals/Slots mode has big overhead and hides too much from developer.

I do not see how it hides anything, it seems to make connections
between things very explicit and clear, while being executing quite
fast and being very simple to use.

On Wed, Jun 2, 2010 at 1:10 AM, Artyom <artyomtnk_at_[hidden]> wrote:
> - HTML and templates are essential for web development, hiding them
> Â behind Qt like "layout" abstracts brings back and not forward.

HTML and templates are like working with the base C API for a
windowing system compared to all of the graphical toolkits built on
top.  Using that low level API gains you *nothing*, since everything
higher can do everything anyway, but the higher levels also help you
to design vastly easier to reuse components and helping to vastly
reduce the amount of bugs while reducing programming and testing time.

On Wed, Jun 2, 2010 at 8:14 AM, David Abrahams <dave_at_[hidden]> wrote:
> At Wed, 2 Jun 2010 00:10:39 -0600,
> OvermindDL1 wrote:
>> I certainly know that is not true when hosting on Windows, on *nix I
>> believe it uses FastCGI. Â The user session data is only important for
>> when the user *has* a session (logged in or so?), otherwise you are
>> free to kill the session once served. Â The amount of data it holds is
>> only your own design, can be as little or large as necessary.
>
> Wow, this Wt thing looks really, really cool. Â I don't suppose they
> want to be part of Boost?

I would doubt it, it is a sub-project of the Emweb company with dual
GPL/Commercial license, they only keep it running and updating to
support their own efforts, like the Qt/Nokia relationship.  Personally
I question some of their design decisions, it would be quite
interesting to create a replacement that is better optimized (in the
boost style) with all that power.  I can propose the question though.

On Wed, Jun 2, 2010 at 8:20 AM, David Abrahams <dave_at_[hidden]> wrote:
> At Wed, 2 Jun 2010 00:10:48 -0700 (PDT),
> Artyom wrote:
>>
>> > Ditto, and a couple years of experience with Wt, and a
>> > decade with
>> > PHP, and a **few months** with CppCMS.
>>
>> Good point.
>>
>>
>> Summary:
>> --------
>>
>> - Developing web application as you develop GUI is not suited for
>> Â  general web development
>
> You haven't given any evidence to support this claim AFAICT, but I may
> have missed it.

Ditto, I have been wondering about that statement too, I have seen no
claims for it, and many against it.

On Wed, Jun 2, 2010 at 8:20 AM, David Abrahams <dave_at_[hidden]> wrote:
>> - Signals/Slots mode has big overhead and hides too much from developer.
>
> I can understand that argument. Â Not sure I agree that it hides too
> much, but at least I understand it.

I am not really sure how it hides anything, I have lots of things
scattered about my code like this:
        userLoggedIn->emit(user);
        clicked().connect(bind(&newPost, this));
        string createImageCall(createImageCallback()+"(type, h, w, text, 'png')");
        {
                JSlot *callOnClient = new JSlot(javascript, this);
                callOnClient.exec("null", "null"); // On the next request to the
client, this will be sent along with all of the other data and
executed
        }
        etc...  This all *vastly* simplifies the code and makes things very explicit

On Wed, Jun 2, 2010 at 8:20 AM, David Abrahams <dave_at_[hidden]> wrote:
>> - HTML and templates are essential for web development, hiding them
>> Â  behind Qt like "layout" abstracts brings back and not forward.
>
> This seems to be an entirely subjective statement. Â I'd like to
> understand whether it's the same as saying “machine instructions are
> essential for coding; hiding them behind Pascal-like ‘high-level
> language’ abstractions brings us back and not forward,” which I would
> reject on its face, or if there's something more to it.
>
> Note that I don't reject all such arguments. Â For example, I believe
> that today's typesafe languages force us too far away from the machine
> model to achieve the highest efficiency. Â But then, that's phrased as
> a tradeoff between safety and efficiency; your statement isn't.

That is exactly why I like the Wt model, it lets you do things in the
high level, in templates, in raw HTML, or anything else you can come
up with, while letting you mix-and-match it all as you wish.