Whiteline: delta/whiteline CD Series - delta.iso/tex/cstex/disk1/cstexdoc.lzh/TEX_DOC/KNUTH.DOC

Text File | 1989-11-02 | 17.5 KB | 337 lines

**********************************************************************
*            The new versions of TeX and Metafont:                   *
*                      Draft description                             *
**********************************************************************
by Donald E. Knuth

For more than five years I held firm to my conviction that a stable
system was far better than a system that continues to evolve. But
during the TUG meeting at Stanford in August, 1989, I was persuaded to
make one last set of changes, in order to bring TeX and Metafont to a
state of completion consistent with their overall philosophy and
goals.

The main reason for the changes was the fact that I had guessed wrong
about 7-bit character sets versus 8-bit character sets. I believed
that standard text input would continue indefinitely to be confined to
at most 128 characters, since I did not think a keyboard with 256
different outputs would be especially efficient. Needless to say, I
was proved wrong, especially by developments in Europe and Asia. As
soon as I realized that a text formatting program with 7-bit input
would rapidly begin to seem as archaic as the 6-bit systems we once
had, I knew that a fundamental revision was necessary.

But the 7-bit assumption pervaded everything, so I needed to take the
programs apart and redo them thoroughly in 8-bit style. This put TeX
onto the operating table and under the knife for the first time since
1984, and I had a final opportunity to include a few new features that
had occurred to me or been suggested by users since then.

The new extensions are entirely upward compatible with previous
versions of TeX and Metafont (with a few small exceptions mentioned
below). This means that error-free inputs to the old TeX and Metafont
will still be error-free inputs to the new systems, and they will
still produce the same outputs.

However, anybody who dares to use the new extensions will be unable to
get the desired results from old versions of TeX and Metafont. I am
therefore asking the TeX community to update all copies of the old
versions as soon as possible. Let us root out and destroy the obsolete
7-bit systems, even though we were able to do many fine things with
them.

In this note I'll discuss the changes, one by one; then I'll describe
the exceptions to upward compatibility.

1. The character set.

   Up to 256 distinct characters are now allowed in input files. The
   codes that were formerly limited to the range 0...127 are now in
   the range 0...255. All characters are alike; you are free to use
   any character for any purpose in TeX, assigning appropriate values
   to its \catcode, \mathcode, \lccode, \uccode, \sfcode, and
   \delcode. Plain TeX initializes these code values for characters
   above 127 just as it initializes the codes for ordinary punctuation
   characters like "!".

   There's a new convention for inputting an arbitrary 8-bit character
   to TeX when you can't necessarily type it: The four consecutive
   characters  ||xy, where x and y are any of the "lowercase
   hexadecimal digits"  0, 1, 2, 3, 4, 5, 6, 7, 8, 9, a, b, c, d, e,
   or f, are treated by TeX on input as if they were a single
   character with specified code digits. For example,  ||80 gives
   character code 128; the entire character set is available from ||00
   to ||ff. The old convention discussed in Appendix C, under which
   character 0 was ||@, character 1 (control--A) was ||A, ..., and
   character 127 was ||?, still works for the first 128 character
   codes, except that the character following || should not be a
   lowercase hexadecimal digit when the immediately following
   character is another such digit.

   The existence of 8-bit characters has less effect in Metafont than
   in TeX, because Metafont's character classes are built in to each
   installation. The normal set of 95 printing characters described on
   page 51 of  Metafontbook can be supplemented by extended characters
   as discussed on page 282, but this is rarely done because it leads
   to problems of portability. Metafont's  char operator is now
   redefined to operate modulo 256 instead  of modulo 128.

2. Hyphenation tables.
   Up to 256 distinct sets of rules for hyphenation are now allowed in
   TeX. There's a new integer parameter called \language, whose
   current value specifies the hyphenation convention in force. If
   \language is negative or greater than 255, TeX acts as if
   \language=0.

   When you list hyphenation exceptions with TeX's  \hyphenation
   primitive, those exceptions apply to the current language only.
   Similarly, the \patterns primitive tells TeX to remember new
   hyphenation patterns for the current language; this operation is
   allowed only in the special "initialization" program called
   INITEX. Hyphenation exceptions can be added at any time, but new
   patterns cannot be added after a paragraph has been typeset.

   When TeX reads the text of a paragraph, it automatically inserts
   "whatsit nodes" into the horizontal list for that paragraph
   whenever a character comes from a different \language than its
   predecessor. In that way TeX can tell what hyphenation rules to use
   on each word of the paragraph even if you switch  frequently back
   and forth among many different languages.

   The special whatsit nodes are inserted automatically in
   unrestricted horizontal mode (i.e., when you are creating a
   paragraph, but not when you are specifying the contents of an
   hbox). You can insert a special whatsit yourself in restricted
   horizontal mode by saying \language<number>. This is needed only if
   you are doing something tricky, like unboxing some contribution to
   a paragraph.

3. Hyphenated fragment control.
   TeX has new parameters  \lefthyphenmin and \righthyphenmin, which
   specify the smallest word fragments that will appear at the
   beginning or end of a word that has been hyphenated. Previously the
   values \lefthyphenmin=2 and \righthyphenmin=3 were hard-wired into
   TeX and impossible to change. Now plain TeX  format supplies the
   old values, which are still recommended for most American
   publications; but you can get more hyphens by decreasing these
   parameters, and you can get fewer hyphens by increasing them. If
   the sum of \lefthyphenmin and \righthyphenmin is 63 or more, all
   hyphenation is suppressed. (You can also suppress  hyphenation by
   using a font with \hyphenchar=-1, or by switching to a  \language
   that has no hyphenation patterns or exceptions.)

4. Smarter ligatures.
   Now here's the most radical change. Previous versions of TeX had
   only one kind of ligature, in which two characters like "f" and "i"
   were changed into a single character like "fi" when they appeared
   consecutively. The new TeX understands much more complex
   constructions by which, for example, we could change an "i"
   following "f" to a dotless "\i" while the "f" remains unchanged:
   "f\i".

   As before, you get ligatures only if they have been provided in the
   font you are using. So let's look at the new features of Metafont
   by which enhanced ligatures can be created. A Metafont programmer
   can specify a "ligature/kerning program" for any character of the
   font being created. If, for example, the "fi" combination appears
   in font position 12, the replacement of "f" and "\i" by "fi" is
   specified by including the statement "i" =: 12 in the
   ligature/kerning program for "f"; this is Metafont's present
   convention.

   The new ligatures allow you to retain one or both of the original
   characters while inserting a new one. Instead of =: you can also
   write |=: if you wish to retain the left character, or =:| if you
   wish to retain the right character, or |=:| if you want to keep
   them both. For example, if the dotless \i appears in font position
   16, you can get the behavior mentioned above by having "i" |=: 16
   in f's program.

   There also are four additional operators |=:>,  =:|>,  |=:|>,
   |=:|>>, where each > tells TeX to shift its focus one position to
   the right. For example, if f and i had been replaced by f and
   dotless \i as above, TeX would begin again to execute f's
   ligature/kern program, possibly inserting a kern before the dotless
   \i, or possibly changing the f to an entirely different character,
   etc. But if the instruction had been "i" |=:> 16 instead, TeX would
   turn immediately to the ligature/kern program for characters
   following character 16 (the dotless \i); no further change would be
   made between f and \i even if the font had something specified
   there.

5. Boundary ligatures.
   Every consecutive string of "characters" read by TeX in horizontal
   mode (after macro expansion) can be called a "word". (Technically
   we consider a "character" in this definition to be either a
   character whose \catcode is a letter or otherchar, or a control
   sequence that has been \let equal to such a character, or a control
   sequence that has been defined by \chardef, or the construction
   \char<number>.) The new TeX now imagines that there is an invisible
   "left boundary character" just before every such word, and an
   invisible "right boundary character" just after it. These boundary
   characters take effect if the font designer has specified ligatures
   and/or kerning between them and the adjacent letters. Thus, the
   first or last character of a word can  now be made to change its
   shape automatically.

   A ligature/kern program for the left boundary character is
   specified within Metafont by using the special label  ||: in a
   ligtable command. A ligature or kern with the right boundary
   character is specified by assigning a value to the new internal
   Metafont parameter  boundarychar, and by specifying a ligature or
   kern with respect to this character. The boundarychar may or may
   not exist as a real character in the font.

   For example, suppose we want to change the first letter of a word
   from "F" to "ff" if we are doing some olde English. The Metafont
   font designer could then say ligtable ||: "F" |:= 11 if character
   11 is the "ff". The same ligtable instruction should appear in the
   programs for characters like ( and ` and " and - that can precede
   strings of letters; then "Bassington-French" will yield
   "Bassington-ffrench".

   If the "s" of our font is the pre-19th century s that looks like a
   mutilated "f", and if we have a modern "s" in position 128, we can
   convert the final s's as Ben Franklin did by introducing ligature
   instructions such as

            boundarychar := 255;
            ligtable "s":   255 =:| 128,
                            "." =:| 128,
                            "," =:| 128,
                            ")" =:| 128,
                            "'" =:| 128,
   and so on. (A true oldstyle font would also have ligatures for ss
   and si and sl and ssi and ssl and st; it would be fun to create a
   Computer Modern Oldstyle.)

   The implicit left boundary character is omitted by TeX if you say
   \noboundary just before the word; the implicit right boundary is
   omitted if you say \noboundary just after it.

6. More compact ligatures.
   Two or more ligtables can now share common code. To do this in
   Metafont, you say "skipto <n>" at the end of one ligtable command,
   then you say "<n>::" within another. Such local labels can be
   reused; e.g., you can say skipto 1 again after 1:: has appeared,
   and this skips to the _next_ appearance of 1::. There are 256 local
   labels, numbered 0 to 255. Restriction: At most 128 ligature or
   kern commands can intervene between a skipto and its matching
   label.

   The TFM file format has been upwardly extended to allow more than
   32,500 ligature/kern commands per font. (Previously there was an
   effective limit of 256.)

7. Better looking sloppiness.
   There is now a better way to avoid overfull boxes, for people who
   don't want to look at their documents to fix unfeasible line breaks
   manually. Previously people tried to do this by setting
   \tolerance=10000, but the result was terrible because TeX would
   tend to consolidate all the badness in one truly horrible line.
   (TeX considers all badness >=10000 to be infinitely bad, and all
   these infinities are equal.)

   The new feature is a dimension parameter called \emergencystretch.
   If \emergencystretch is positive and if TeX has been unable to
   typeset a paragraph without exceeding the given tolerances, another
   pass over the paragraph is made in which TeX pretends that
   additional stretchability equal to \emergencystretch is present in
   every line. The effect of this is to scale down all the badnesses
   into a range where previously infinite cases become finite;  TeX
   will find an optimum solution to the scaled-down problem, and this
   will be about as good as possible in a practical sense. (The extra
   stretching is not really present; therefore underfull boxes will be
   reported in warning messges unless \hbadness is increased.)

8. Looking at badness.
   TeX has a new internal integer parameter called \badness that
   records the badness of the box it has most recently constructed. If
   that box was overfull, \badness will be 1000000; otherwise \badness
   will be between 0 and 10000.

9. Looking at the line number.
   TeX also has a new internal integer parameter called \inputlineno,
   which contains the number of the line that TeX would show on an
   error message if an error occurred now. (This parameter and
   \badness are "read only" in the same way as \lastpenalty: You can
   use them in the context of a <number>, e.g., by saying
   "\ifnum\inputlineno>\badness ...\ \fi" or "\the\inputlineno", but
   you cannot set them to new values.)

10. Not looking at error context.
   There's a new integer parameter called \errorcontextlines that
   specifies the maximum number of two-line pairs of context displayed
   with TeX's error messages (in addition to the top and bottom lines,
   which always appear). Plain TeX now sets \errorcontextlines=5, but
   higher level format packages might prefer \errorcontextlines=1  or
   even \errorcontextlines=0. In the latter case, an error that
   previously involved three or more pairs of context would now appear
   as follows:
            ! Error.
            <somewhere> The \top
                        The \top\ line
            ...
            1.123 \The
                  \The\ bottom line.
   (If \errorcontextlines<0 you wouldn't even see the `...' here.)

11. Output recycling.
   One more new integer parameter completes the set. If
   \holdinginserts>0 when TeX is putting the current page into \box255
   for the  \output routine, TeX will not move anything from insertion
   nodes into the corresponding boxes; all insertion nodes will stay
   in place. Designers of output routines can use this when they want
   to put the contents of box 255 back into the current page to be
   re-broken (because they might want to change \vsize or something).

12. Exceptions to upward compatibility.
   The new features of TeX and Metafont imply that a few things work
   differently than before. I will try to list all such cases here
   (except when the  previous behavior was erroneous due to a bug in
   TeX or Metafont). I don't know of any cases where users will
   actually be affected, because all of these exceptions are pretty
   esoteric.

   o TeX used to convert the character strings ||0, ||1,..., ||9, ||a,
     ||b, ||c, ||d, ||e, ||f into the respective single characters p,
     q,..., y, !, ", #, $, %, &. It will no longer do this if the
     following character is one of the characters 0123456789abcdef.

   o TeX used to insert no character at the end of an input line if
     \endlinechar>127. It will now insert a character unless
     \endlinechar>255. (As previously, \endlinechar<0 suppresses the
     end-of-line character. This character is normally 13=ASCII
     control--M = carriage return.)

   o Some diagnostic messages from TeX used to have the notation
     ["80]...["FF] when referring to characters 128...255 (for example
     when displaying the contents of an overfull box involving fonts
     that include such characters). The notation ||80...||ff is now
     used instead.

   o The expressions char128 and char0 used to be equivalent in
     Metafont; now char is defined modulo 256 instead. Hence char-1 =
     char255, etc.

   o INITEX used to forget all previous hyphenation patterns each time
     you specified \patterns. Now all hyphenation pattern specifications
     are cummulative, and you are not permitted to use \patterns after a
     paragraph has been hyphenated by INITEX.

   o TeX used to act a bit differently when you tried to typeset
     missing characters of a font. A missing character is now considered
     to be a word boundary, so you will get slightly more diagnostic
     output when \tracingcommands>0.

   o TeX and Metafont will report different statistics at the end of a
     run because they now have a different number of primitives.

   o Programs that use the string pool feature of TANGLE will no
     longer run without changes, because the new TANGLE starts numbering
     multicharacter strings at 256 instead of 128.

   o INITEX programs must now set \lefthyphenmin=2 and
     \righthyphenmin=3 in order to reproduce their previous behavior.

ə