[plt-dev] Fwd: [plt-scheme] How to add a new language to DrScheme?

On Wed, Feb 18, 2009 at 2:10 PM, Eli Barzilay <eli at barzilay.org> wrote:
> On Feb 18, Carl Eastlund wrote:
>>
>> Consider the web server as an example.  We do not provide code for
>> the web server that defines its configuration values and tell the
>> user to "edit collects/web-server/config.ss".  Instead, we provide a
>> framework for the web server that is parameterized over its
>> configuration.  The user can then write a program or run a command
>> line utility that fills in those parameters and runs the web server.
>> We provide the entire framework; the user writes one or a few lines
>> to kick it off, rather than editing files in our provided collection
>> tree.
>
> To create simple a web-server application you write some code in a
> file and package it up in some form that your clients can install.  To
> create a basic customized handin client you write some code in a file
> (and it's very simple: just an info.ss file) and you need to package
> it in a way your clients can install.  IOW, I don't see any point
> here.
>
> It just happens that the *actual* client code (the one in the
> handin-client collection) needs to be packaged with it right now --
> but that won't be necessary if that code is included with the
> student's drscheme installation.  To take this analogy back to the
> web server -- if it was not included with plt, then to write a web
> server application you'd need to package your code and the web-server
> code and ship all of that to your clients -- same deal.
>
> Some more details: I think that it's safe to assume that I'm using the
> handin server/client pretty heavily -- so this is a version that is
> supposedly going to be an extreme case of "customization".
>
> * What I do to setup the server is run it from a directory with a
>  configuration file and some additional things (like the ssl
>  certificate files, the hw directories etc).  The server code itself
>  is _unmodified_ (it's actually running from the plt svn tree).
>
> * What I do to setup the client is:
>
>  1. copy the whole thing over to my own directory
>
>  2. overwrite server-cert.pem with my own certificate
>
>  3. overwrite icon.png with my own icon
>
>  4. edit info.ss
>
>  5. add a whole bunch of stuff that is specific to my course
>
>  Now, if the handin client was included with the student
>  distributions,
>
>  - I'd begin with an empty directory
>
>  - I wouldn't need to do #1
>
>  - In #2 and #3 I'd "put my files" in instead of "overwriting" them.
>
>  - For #4 I'd "write my own" info file, where this is most
>    conveniently done by starting from a documented example that lists
>    what I can have there -- a template file happens to be a very
>    convenient format for such work (the template could move from a
>    file to the manual, but it's the same idea)
>
>  - #5 would be the same
>
>  - Additional point: with the way things are, I'd probably need to
>    write stub files like a "client-gui.ss" file that has:
>
>      #lang scheme/base
>      (require handin-client/client-gui)
>      (provide (all-from-out handin-client/client-gui))
>
>    But if the collection is included with plt, then it should be easy
>    to arrange it so the stub files are not needed.  That is, the info
>    file will use the tool directly from the plt tree, and things will
>    work fine.  (The main change here would be in the
>    "this-collection.ss" hack, which will find the collection holding
>    the info file instead of itself.)
>
> If my guess is correct, then you're assuming that I won't do that last
> paragraph and that's what you're trying to complain about.  This
> assumption is false.

It sounds like we agree about the kind of changes to make to the
handin server/client distribution process.

-- 
Carl Eastlund