Notes on the Gwydion Dylan Bindings to PCRE
Written by Tom Emerson, tree@tiac.net
Revision 0.1

Overview
--------

PCRE (Perl-compatible Regular Expressions) is a C library written by
Philip Hazel (ph10@cam.ac.uk) which provides Perl 5.004 and 5.005
compatible regexp functionality. The Python scripting language uses
PCRE as its regular expression engine.

This is the first go at a set of Gwydion Dylan bindings for PCRE ---
the eventual goal being to replace the existing regexp library with
PCRE (while maintaining the interface, of course). The mappings are
spartan and need to be improved, but I wanted to get this out for
people to start using while I muck with cleaning up the Dylan
wrappers.

Included in this distribution are the following:

GDREADME                  this file
gd-makefile.gmk           makefile for building the Dylan library
pcre-exports.dylan        library and module definitions
pcre-intr.intr            Melange interface definitions
pcre.lid                  LID file for the library
pcretest.patch            patch to the pcretest program (see below)
gd-pcretest/              Dylan implementation of the pcretest program
gd-pgrep/                 Dylan implementation of the pgrep program


Building
--------

1. Build PCRE according to its directions (pretty much just type
   'make')

2. Build the GD libraries with, 'make -f gd-makefile.gmk'.

I haven't bothered to create makefiles for gd-pcretest and gd-pgrep
yet: both can be compiled by dropping into their directory and typing

d2c -L.. <lid-file>

gd-pcretest works very much like its C and Perl counterparts. It takes
two arguments, the name of the input file containing test data, and
the name of the output file to which the results are written.

Unfortunately the "%x" specifier to 'format' doesn't have the same
semantics as the C version: to make it easy to compare the Dylan test
results with those for the C implementation, I made a slight
modification to pcretest.c to output hex numbers in the same format as
GD's 'format'. Just apply 'pcretest.patch' to pcretest.c and
rebuild. Then regenerate the test output for 'testinput' and
'testinput3':

% ./pcretest testinput testoutput.gd
% ./pcretest testinput3 testoutput3.gd

Given this, after running gd-pcretest the output for testinput and
testinput3 should be identical.

Note that testinput2 and testinput4 test features of PCRE that I am
not currently supporting.

The performance of gd-pcretest leaves a lot to be desired: I was not
careful about minimizing the number of string copies that are
performed so I'm sure there is much room for improvement. However,
since the purpose is to simply test the operation of the interface I'm
not concerned much about the performance.

The gd-pgrep utility behaves almost exactly like its C counterpart:
the only differences are:

- *standard-input* is output instead of 'stdin'.
- Return values don't match (this will be fixed later)


Future Directions
-----------------

- Improve error handling: the nonsense of expecting the user to deal
  with error strings and whatnot is silly. Conditions should be used
  instead.

- Harlequin C-FFI version. Perhaps as a test of Pidgin to compare with
  a hand-generated interface.

- Write extraction functions to extract matches from the search
  string.

- Fix bugs and improve my Dylan style.

- Investigate i18n issues in the engine.


1999-Mar-30
tree@tiac.net