The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
=head1 NAME

bt_postprocess - post-processing of BibTeX strings, values, and entries

=head1 SYNOPSIS

   void bt_postprocess_string (char * s,
                               btshort options)

   char * bt_postprocess_value (AST *   value,
                                btshort  options, 
                                boolean replace);

   char * bt_postprocess_field (AST *   field, 
                                btshort  options, 
                                boolean replace);

   void bt_postprocess_entry (AST *  entry,
                              btshort options);

=head1 DESCRIPTION

When B<btparse> parses a BibTeX entry, it initially stores the results
in an abstract syntax tree (AST), in a form exactly mirroring the parsed
data.  For example, the entry

   @Article{Jones:1997a,
     AuThOr = "Bob   Jones" # and # "Jim Smith ",
     TITLE = "Feeding Habits of
              the Common Cockroach",
     JoUrNaL = j_ent,
     YEAR = 1997
   }

would parse to an AST that could be represented as follows:

   (entry,"Article")
     (key,"Jones:1997a")
     (field,"AuThOr")
       (string,"Bob   Jones")
       (macro,"and")
       (string,"Jim Smith ")
     (field,"TITLE")
       (string,"Feeding Habits of               the Common Cockroach")
     (field,"JoUrNaL")
       (macro,"j_ent")
     (field,"YEAR")
       (number,"1997")

The advantage of this form is that all the important information in the
entry is readily available by traversing the tree using the functions
described in L<bt_traversal>.  This obvious problem is that the data is
a little too raw to be immediately useful: entry types and field names
are inconsistently capitalized, strings are full of unwanted whitespace,
field values not reduced to single strings, and so forth.

All of these problems are addressed by B<btparse>'s post-processing
functions, described here.  Normally, you won't have to call these
functions---the library does the Right Thing for you after parsing each
entry, and you can customize what exactly the Right Thing is for your
application.  (For instance, you can tell it to expand macros, but not
to concatenate substrings together.)  However, it's conceivable that you
might wish to move the post-processing into your own code and out of the
library's control.  More likely, you could have strings that come from
something other than BibTeX files that you would like to have treated as
BibTeX strings; for that situation, the post-processing functions are
essential.  Finally, you might just be curious about what exactly
happens to your data after it's parsed.  If so, you've come to the right
place for excruciatingly detailed explanations.

=head1 FUNCTIONS

B<btparse> offers four points of entry to its post-processing code.  Of
these, probably only the first and last---for processing individual
strings and whole entries---will be commonly used.

=head2 Post-processing entry points

To understand why four entry points are offered, an explanation of the
sample AST shown above will help.  First of all, the whole entry is
represented by the C<(entry,"Article")> node; this node has the entry
key and all its field/value pairs as children.  Entry nodes are returned
by C<bt_parse_entry()> and C<bt_parse_entry_s()> (see L<bt_input>) as
well as C<bt_next_entry()> (which traverses a list of entries returned
from C<bt_parse_file()>---see L<bt_traversal>).  Whole entries may be
post-processed with C<bt_postprocess_entry()>.

You may also need to post-process a single field, or just the value
associated with it.  (The difference is that processing the field can
change the field name---e.g. to lowercase---in addition to the field
value.)  The C<(field,"AuThOr")> node above is an example of a field
sub-AST, and C<(string,"Bob   Jones")> is the first node in the list of
simple values representing that field's value.  (Recall that a field
value is, in general, a list of simple values.)  Field nodes are
returned by C<bt_next_field()>, value nodes by C<bt_next_value()>.  The
former may be passed to C<bt_postprocess_field()> for post-processing,
the latter to C<bt_postprocess_value()>.

Finally, individual strings may wander into your program from many
places other than a B<btparse> AST.  For that reason,
C<bt_postprocess_string()> is available for post-processing arbitrary
strings.

=head2 Post-processing options

All of the post-processing routines have an C<options> parameter, which
you can use to fine-tune the post-processing.  (This is just like the
per-metatype string-processing options that you can set before parsing
entries; see C<bt_set_stringopts()> in L<bt_input>.)  Like elsewhere in
the library, C<options> is a bitmap constructed by or'ing together
various predefined constants.  These constants and their effects are
documented in L<btparse/"String processing option macros">.

=over 4

=item bt_postprocess_string ()

   void bt_postprocess_string (char * s,
                               btshort options)

Post-processes an individual string, C<s>, which is modified in place.
The only post-processing option that makes sense on individual strings
is whether to collapse whitespace according to the BibTeX rules; thus,
if C<options & BTO_COLLAPSE> is false, this function has no effect.
(Although it makes a complete pass over the string anyways.  This is for
future expansion.)

The exact rules for collapsing whitespace are simple: non-space
whitespace characters (tabs and newlines mainly) are converted to space,
any strings of more than one space within are collapsed to a single
space, and any leading or trailing spaces are deleted.  (Ensuring that
all whitespace is spaces is actually done by B<btparse>'s lexical
scanner, so strings in B<btparse> ASTs will never have whitespace apart
from space.  Likewise, any strings passed to bt_postprocess_string()
should not contain non-space whitespace characters.)

=item bt_postprocess_value ()

   char * bt_postprocess_value (AST *   value,
                                btshort  options, 
                                boolean replace);

Post-processes a single field value, which is the head of a list of
simple values as returned by C<bt_next_value()>.  All of the relevant
string-processing options come into play here: conversion of numbers to
strings (C<BTO_CONVERT>), macro expansion (C<BTO_EXPAND>), collapsing of
whitespace (C<BTO_COLLAPSE>), and string pasting (C<BTO_PASTE>).  Since
pasting substrings together without first expanding macros and
converting numbers would be nonsensical, attempting to do so is a fatal
error.

If C<replace> is true, then the list headed by C<value> will be replaced
by a list representing the processed value.  That is, if string pasting
is turned on (C<options & BTO_PASTE> is true), then this list will be
collapsed to a single node containing the single string that results
from pasting together all the substrings.  If string pasting is not on,
then each node in the list will be left intact, but will have its
text replaced by processed text.

If C<replace> is false, then a new string will be built on the fly and
returned by the function.  Note that if pasting is not on in this case,
you will only get the last string in the list.  (It doesn't really make
a lot of sense to post-process a value without pasting unless you're
replacing it with the new value, though.)

Returns the string that resulted from processing the whole value, which
only makes sense if pasting was on or there was only one value in the
list.  If a multiple-value list was processed without pasting, the last
string in the list is returned (after processing).

Consider what might be done to the value of the C<author> field in the
above example, which is the concatenation of a string, a macro, and
another string.  Assume that the macro C<and> expands to C<" and ">, and
that the variable C<value> points to the sub-AST for this value.
The original sub-AST corresponding to this value is

   (string,"Bob   Jones")
   (macro,"and")
   (string,"Jim Smith ")

To fully process this value in-place, you would call

   bt_postprocess_value (value, BTO_FULL, TRUE);

This would convert the value to a single-element list,

   (string,"Bob Jones and Jim Smith")

and return the fully-processed string C<"Bob Jones and Jim Smith">.
Note that the C<and> macro has been expanded, interpolated between the
two literal strings, everything pasted together, and finally whitespace
collapsed.  (Collapsing whitespace before concatenating the strings
would be a bad idea.)

(Incidentally, C<BTO_FULL> is just a macro for the combination of all
possible string-processing options, currently:

   BTO_CONVERT | BTO_EXPAND | BTO_PASTE | BTO_COLLAPSE

There are two other similar shortcut macros: C<BTO_MACRO> to express the
special string-processing done on macro values, which is the same as
C<BTO_FULL> except for the absence of C<BTO_COLLAPSE>; and
C<BTO_MINIMAL>, which means no string-processing is to be done.)

Let's say you'd rather preserve the list nature of the value, while
expanding macros and converting any numbers to strings.  (This
conversion is trivial: it just changes the type of the node from
C<BTAST_NUMBER> to C<BTAST_STRING>.  "Number" values are always stored
as a string of digits, just as they appear in the file.)  This would be
done with the call

   bt_postprocess_value
      (value, BTO_CONVERT|BTO_EXPAND|BTO_COLLAPSE,TRUE);

which would change the list to

   (string,"Bob Jones")
   (string,"and")
   (string,"Jim Smith")

Note that whitespace is collapsed here I<before> any concatenation can
be done; this is probably a bad idea.  But you can do it if you wish.
(If you get any ideas about cooking up your own value post-processing
scheme by doing it in little steps like this, take a look at the source
to C<bt_postprocess_value()>; it should dissuade you from such a
venture.)

=item bt_postprocess_field ()

   char * bt_postprocess_field (AST *   field, 
                                btshort  options, 
                                boolean replace);

This is little more than a front-end to C<bt_postprocess_value()>; the
only difference is that you pass it a "field" AST node (eg. the
C<(field,"AuThOr")> in the above example), and that it transforms the
field name in addition to its value.  In particular, the field name is
forced to lowercase; this behaviour is (currently) not optional.

Returns the string returned by C<bt_postprocess_value()>.

=item bt_postprocess_entry ()

   void bt_postprocess_entry (AST *  entry,
                              btshort options);

Post-processes all values in an entry.  If C<entry> points to the AST
for a "regular" or "macro definition" entry, then the values are just
what you'd expect: everything on the right-hand side of a field or macro
"assignment."  You can also post-process comment and preamble entries,
though.  Comment entries are essentially one big string, so only
whitespace collapsing makes sense on them.  Preambles may have multiple
strings pasted together, so all the string-processing options apply to
them.  (And there's nothing to prevent you from using macros in a
preamble.)

=back

=head1 SEE ALSO

L<btparse>, L<bt_input>, L<bt_traversal>

=head1 AUTHOR

Greg Ward <gward@python.net>