Hello people,
sorry  if  these  issues  were discussed before, but i see no trace of
them being implemented so i assume they were not.

i  was  thinking what a decent (imho sure) appserver needs to have and
here are the results:

3-domains: i think it's a nice feature to have: imagine that there are
3  levels of abstraction that are as follows: address space (there are
actually  should  be  as  many address spaces as there are protocols),
virtual  space  (this  is  a  space, servlets operate in) and physical
space.

So  if the url prot://domain/path/dir is called, Application looks for
an address space (AS) handler for a protocol 'prot' that handler looks
at  a  domain name and decides what AS handler will be used on it that
handler in turn looks at the path (it probably should be informed that
the  protocol and domain parts are already processed) and decides what
virtual  space  handler  should be used on it (or direct processing of
the  AS name further to other handlers), virtual space (VS) name is to
look  something  like 'foo.bar.name', here it's processed in a similar
way and in the end is transformed into a file name plus in a case of a
servlet a class name (that has to be implemented in that file), then a
logic similar to the one present i nwebkit now comes to play.

this  gives us many advantages. we can easily implement an (incredibly
flexible)  aquisition  (for  people  who  want  it) in a virtual space
(that's  a lot different from zope-way (russian speakers know what i'm
talking   about)   aquisition   that  exposes  all  the  guts  of  the
application), we can easily control what is published and what is not,
we  can allow for a much more custom handling of requests (things that
are  done  with  extrapathinfo  could  be done in a more natural way),
there  could  be  as many servlets per file as desired, you can simply
implement  virtual  hosting  and  things  like  that,  there  are some
benefits for caching (same servlet on multiple urls) and more things i
can't remember, and it's 100% backwards compatible!

middlekit:  why  not to save a model that was used to generate sql and
.py  stubs?  if  we  do  we  can see what changes were made since last
generation  and  generate  an  'alter'  statements  in  sql instead of
dropping  and recreating tables (we can fill new fields of old records
with  default  values specified in a dedicated column or NULLs if none
are  specified  and  NULL is permitted). this would enable us to alter
models of production phase sites and abandon Samples.csv files.

debugging:  there should be hooks for a programmer to determine if the
servlet is used in debug mode (so 'user' determines it and webkit uses
that  info) and proceed in some way like this: if debugging reload all
changed  modules;  if  failed  (unhandled  exception  in servlet): and
debugging  generate  tracebacks  and  display  them,  else  just  send
traceback  to  webmaster  and  show some error message to user without
exposing  any  source  and  whatever  else.  debugging  mode should be
possible  to  set explicitly, or be detected by the host of the client
(localhost  is  likely  to be debugging) and by custom logic. and then
after  deployment  debug  mode could be locked out completely.
i believe you should see the idea by this time.

i  hope  all  of  this is not completely stupid and some comments will
follow.

regards
-- 
Serge Shchetinin a.k.a. Maluke.
3.Apr.2002


_______________________________________________
Webware-discuss mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/webware-discuss

Reply via email to