[Ecls-list] CLOS streams

[Juan Jose Garcia-Ripoll]

Since Beppe's version of ECL, this lisp interpreter has supported the 
possibility of using user-defined objects as input/output streams. The
code has been reworked so that it is easier to maintain. Nevertheless
I post the following announcement so that people become aware of this
feature. I welcome any comments directed to improving this interface.

* System design:

  - Global function pointer output_ch_fun and read_ch_fun
    disappear. Now the differences between ordinary streams and CLOS
    streams are handled in file.d. Optimizations for synonym streams
    also disappear from read.d and print.d.

* Visible changes and ANSI compatibility:

  - New configuration flag --with-clos-streams, enable the user to
    define CLOS objects which behave as lisp streams. Such object must
    have the following methods defined: STREAM-READ-CHAR,
    STREAM-UNREAD-CHAR, STREAM-LISTEN, STREAM-CLEAR-INPUT,
    STREAM-WRITE-CHAR, STREAM-CLEAR-OUTPUT, STREAM-FORCE-OUTPUT,
    STREAM-CLOSE.

User defined streams are very straightforward to implement (See previous posts 
from Andrew Topp in this list), but they can be rather slow, since 
input/output is performed character-wise, without buffering.

The changes I have made are related to removing many "optimizations" in the 
PRINT/READ routines. This code worked checked the input of PRINT, READ, etc, 
and when it was a CLOS object, it invoked one of the method mentioned above. 
This was tricky because the type of the stream and the routines to handle it 
had to be kept in global pointers, which gave rise to some errors already 
discussed in this list.

What I have done is to move all this code to file.d. When the user compiles 
ECL using --with-clos-streams, all low level routines which read/write to 
streams (not PRINT, READ, etc, but lower level functions) accept both 
ordinary streams and user defined objects. This shouldn't slow the operation 
with ordinary streams much, and it also makes the code more robust, since all 
these low level routines now check the type of its arguments.

Juanjo

-- 
Max-Planck-Institut fuer Quantenoptik	+49/089/32905-127
Hans-Kopfermann-Str. 1, D-85748		www.arrakis.es/~worm
Garching b. Muenchen, Germany		worm at arrakis.es