Re[4]: Is Smart Eiffel ready for large-scale application ?

Hello Marcio,

Saturday, April 9, 2005, 7:01:18 PM, you wrote:

MM>  Using their free version just in the development phase (because you get incremental
MM> compilation, debugging etc) and SE for deployment does not work well because of the
MM> differences in both implementations (even if you use GOBO end EPOSIX). As you said,
MM> Thread etc is still problematic, plus even basic language features. Maybe if we had a
MM> standard and implementors followed it, but I don't feel optimistic about it. In some
MM> sense, it feels as if Eiffel was being invented now, with discrepancies across
MM> implementations because of a lack of agreement at the early phases. Again, maybe in a
MM> few years, with standards... Maybe...

I don't see any sign why this should happen.
In my eyes everything is getting worse instead of better.

MM>  Personally I think that Lothar's project is a great test case for the language &
MM> libraries & compiler. It is not some batch-mode program
MM> (compile this into that), but a 
MM> fully interactive application that has to be competitive with state-of-the-art. Not just
MM> a command-line tool or some Mickey Mouse GUI quickly put
MM> together by command-line lovers 
MM> (Linux is full of these). I strongly believe that you have to "eat your own dog food" as
MM> they say (SE team using SE to implement the compiler, etc), but you also need a
MM> few "real customers" in order to "keep you honest" in terms of your assessment of what
MM> really matters and how far off you are.

Yes. For all those who don't know what we are talking about. It's
"http://www.ruby-ide.com".  And i got around 10000 downloads during
the last year and at least two dozen person who use it already every
day and actively reporting bugs/comments.

So the first crucial test i done.

MM> for many other languages too !!!). Again, Lothar's IDE is a great test case for the
MM> tools used, so I truly respect his insights. I've had a similar SE experience myself,
MM> although I was careful to "fail early" in my evaluation of the tool (started with the
MM> risky parts first, and saw how far I could push with SE). It's been years now, and the
MM> problems are still there. Sigh. For now, I keep wathcing...

Unfortunately it is completely different. If i would try to track down
bugs like Dominique, Cyril, Philippe or any of the other team members
i would be already out of business. Smalleiffel and SmartEiffel (i got
it compilable on 2.1) crashs just to many times. Luckily i soon found out
which constructs i must avoid so it happens not so often any more.

I mean it's really stupid. It simply doesn't matter if a test case
that crashs is 6 MB large or 25 lines. If you can put your finger on
the line and tell you change this 'a' into 'b' and compile_to_c
crashes it's the best bug report you can get from a user. With
DbC the size of the testcase simply does not matter. This is something
i learned very fast. And even if the crash report is not so
reproduceable, does the team add special DbC and work with someone to
track it down? No! The usual arrogance is shown and everything
is rejected.

That's why i can't take the team as serious anymore. They have this
stupid "it works for me, so shut up and don't steal my time" attitude.
SmartEiffel will never get very far if they don't change this. For
example i would consider the whole "net" cluster as useless or at
least very bad. Starting with implementing it posix like and then
trying to port it to windows instead of asking for help in design phase.
They don't see that the select model
and even there polling model only works in a few cases and on unix.
GUI's (especially Win32) don't work very well with it and it's
fucking slow and so old fashioned compared to the technical possible
solution that we could have with agent callback functions on
windows if it were designed well.

It's a pitty for the eiffel community (is there still anyone ?). The
smarteiffel project is killing the public opinion of the language.
Just because they don't care about practical issues and prefer
a theorectical better system (even this i would doubt). But as long as
SmartEiffel is alive there will be nobody working on another open eiffel
system. I did not follow the Visual Eiffel project, but it's so quite
about this on comp.lang.eiffel that i don't expect very much.

By the way is anybody out there using SmartEiffel > 2.1 for anything
serious ? Something more then a small script project ? I really doubt
it. The fact that we still don't have GOBO,EPOSIX or any new interesting
SmartEiffel project is a clear indicator of it.


And a final note: I love it to write a small component for my IDE with
smalleiffel and then integrate it into my system. Giving me a feeling
how good the whole thing could be if we were working together.





-- 
Best regards,
 Lothar                            mailto:[email protected]