COMMENT ⊗   VALID 00025 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00005 00002	This file describes differences between TeX82, which is the portable standard
C00010 00003	* Characters by themselves can no longer masquerade as numbers. Thus, you
C00014 00004	* New input capabilities greatly expand the class of potential applications:
C00018 00005	* \chpar has been abolished. In its place, all of the integer parameters have
C00022 00006	* The \tracing parameter disappears, and its various components each have their
C00027 00007	* An array of 256 dimension values is introduced, called \dimen0 to \dimen255.
C00030 00008	* If you put \global in front of \def or \let or \chcode or \tolerance or
C00034 00009	* If you say \message{text}, the terminal will display " text" immediately.  For
C00039 00010	* The internal character set used by TeX82 is the same regardless of the
C00049 00011	* Note that the backslash character is now predefined as an escape character
C00055 00012	* \indent takes you from vertical mode to horizontal mode and indents the
C00060 00013	* Up to 256 fonts may be used, and each font may contain up to 256 characters.
C00065 00014	* TeX82 has a new "help" facility available on error messages. If you type "h"
C00070 00015	* The \mathcode now has a 15-bit number as its value. The first three bits
C00076 00016	* The spacing in math formulas is unbundled too there are three parameters
C00081 00017	* The maximum penalty has been raised from 1000 to 10000 thus, "\penalty 1000"
C00086 00018	* Major changes have been made to the \output conventions, so that it will be
C00101 00019	* In vertical mode, you can say \prevdepth=3pt to make TeX82 act as if the
C00106 00020	* Characters chcoded 13 are not equivalent to one-letter control sequences. They
C00111 00021	* \hbox par is eliminated! Instead of "\hbox par 100pt{...}", one now says
C00116 00022	* \accent is allowed only in horizontal mode \mathaccent only in math mode.
C00122 00023	* Here is an extension to the language intended to placate people who
C00126 00024	* Syntactic conditionals!
C00130 00025	* \everymath{...} inserts its tokens into TeX's scanner just when
C00135 ENDMK
C⊗;
This file describes differences between TeX82, which is the portable standard
definition of TeX in PASCAL, and the TeX systems written in SAIL.  (The SAIL
version will NOT be brought up to date to make it compatible with TeX82; its use
should gradually die away as more people take advantage of the new features
available in the PASCAL version.)

* TeX82 does all its calculations that affect line breaking and page breaking
using fixed-point integer arithmetic of limited (i.e., 32-bit) precision,
instead of with floating-point computations, since different machines differ so
widely in the results you get with floating point.

Dimensions are integers in units of 2↑(-16) points, limited in magnitude to 2↑14
points (which is 18.89 feet).  This applies to all dimensions (e.g., the heights
and widths and depths of boxes, the amounts by which you \raise or \lower a box,
\varunits, etc.), except for the dimensions of stretching and shrinking.

Something different had to be done with respect to the dimensions of stretching
and shrinking, since for example the old TeX defined \hfill to be a stretch of
10↑10 points, and that number has more than 32 bits to the left of its binary
point. After considering various alternatives, the solution introduced by the
designers of MESA-TeX in 1979 has been adopted for TeX82. Each stretch or shrink
dimension is specified by a fixed point integer that is either in units of
2↑(-16)pt or 2↑(-16)fil or 2↑(-16)fill or 2↑(-16)filll.  Here pt is, of course,
one point; the other units are three orders of infinity, essentially infinity
and infinity↑2 and infinity↑3. To add together such units of stretching or
shrinking, one simply adds the individual components having the same order of
infinity, and then uses the nonzero component having the highest order.

Thus, when one says "\hskip <a>pt plus <b>fil minus <c>filll", the numbers <a>,
<b>, <c> are rounded to the nearest multiple of 2↑(-16), and their magnitudes
should be less than 2↑(15). The stretch component is <b> times infinity, and the
shrink component is <c> times infinity cubed.

* The shrink component of all glue used in a paragraph should be finite.
Something like "\hskip 20pt minus 2fil" actually makes no sense in a paragraph,
since the paragraph would fit on a single line no matter what. Infinite
shrinkage does make sense in a simple \hbox, of course.

* Identifiers for control sequences in TeX82 are letter strings of any length,
with upper and lower case letters treated as distinct even when they aren't the
first letter. For example, \TeX is not the same as \TEX, and \GAMMA is not the
same as \Gamma.  Any character that is regarded as a letter (this means the 52
letters, initially, plus others that are \chcoded to 11) can appear in such
control sequences. Of course, the one-character non-letter control sequences
still exist as well.
* Characters by themselves can no longer masquerade as numbers. Thus, you
shouldn't say "\chcode A" any more; this has caused more mysterious errors than
it was worth, and anyway there will be a macro for defining fonts symbolically
in TeX82. Instead, one can use a left quote in front of a character to get its
ascii code; e.g., `A is '101, and `↑↑A is 1.  Furthermore you can use an escape
in front of the character, e.g., `\A or `\% (these macros will not be expanded),
so that we avoid the problem in TeX82 that you couldn't say "\chcode %" when %
had already been chcoded.

* Hexadecimal constants are allowed, using a prefixed ". For example, "DF = '337
= 223.

* \uccode <char number> is the character code to use when converting to upper
case in the \uppercase function. For example, \uccode`a =`A.  (Probably nobody
will change the default values unless a foreign alphabet is being used.)
Characters whose uccode is zero will not be changed by the \uppercase function.

* \lccode <char number> is the character code to use when converting to lower
case in the \lowercase function or when trying to hyphenate a word that contains
upper case letters. For example, \lccode`A =`a. Characters whose lccode is zero
will not be changed by the \lowercase function.  The lccode is also used for
hyphenation; a character whose lccode is zero will be regarded as a nonletter
(e.g., punctuation), and not part of a hyphenatable word, while a character
whose lccode is identical to the character itself will be considered a lower
case letter. For example, \lccode`a =`a.

* If \uchyph is nonzero, a word whose first letter is upper case will be subject
to hyphenation. (This means a word whose first letter has lccode nonzero and
lccode not equal to the character itself.) Words whose first letter is lower
case will always be subject to hyphenation even if they contain upper case
letters further on.

* Both \looseness and \parshape are now reset to their default values after each
paragraph, as \hangindent always was.

* Font codes are now arbitrary control sequences instead of letters, so there is
no longer a 64-font restriction. For example, one says "\font\f=cmr10 at 12pt"
to load the font identified by \f. A subsequent appearance of the control
sequence \f will select this font. (The \: control sequence is no longer
present in TeX82.)
* New input capabilities greatly expand the class of potential applications:

\openin n=filename, where the = is optional and where the stream number n
is between 0 and 15, designates a text file that can be read
concurrently with other input. If a file has already been opened for
the same number n, it is closed first.

\closein n simply closes stream n.

\ifeof n {...} \else {...} tests if input stream n is either not open
or has been fully read.

\read n \cs and \global\read n \cs: These are the important new commands.
They define control sequence \cs to be the contents of the next line from
input stream n, reading that line in the normal way: the current chcodes
are used, spaces are ignored at the beginning of the line, a blank line
comes through as \par, etc. If input stream n has not been opened, or if
an opened file has been fully input, a line is read from the terminal
instead; the user is prompted "\cs=" in this case. The latter feature
allows convenient interactive routines, e.g.,
	\message{Please type your name:}
	\read 0 \myname
	\message{Hello, \myname!}

* The previous commands \open and \send are renamed \openout and \write,
to show their similarity to \openin and \read.

\closeout <number> will close a file so that your TeX program can immediately
input it. (Previously you could only do this by, e.g., \openout 0=empty0.tmp,
cluttering up the disk with an empty file.)

\write, \openout, and \closeout will be ignored if they occur within a \leaders
construction. (Reason: The number of times a leader box occurs might be 0 or 1
depending on floating-point rounding. This restriction keeps the language
independent of floating point.)

* \ifx now allows comparison of any two tokens, not just control sequences.
A non-control-sequence matches only another non-control-sequence that
has the same chcode and represents the same character.

* New feature \expandafter t, where t is any token (usually a control sequence):
If the token following t is a macro, it is expanded as if t were not there.
Then t is put back in front of the result. For example,
	\def\a{\x\y\z} \def\b#1\z{#1} \expandafter\b\a
yields \b\x\y\z which yields \x\y. (It used to be possible to do this only
with trickery/hackery.)
* \chpar has been abolished. In its place, all of the integer parameters have
names instead of numbers. (Thus at last they become consistent with the
dimension parameters \hsize, etc., and with the glue parameters \baselineskip,
etc. My only excuse for bad design in the first place was that the integer
parameters were afterthoughts, stuck in after TeX was first up and running; it
was the easiest way to vary some of the originally fixed constants. I wanted to
finish TeX in a year and get on to writing Volume 4! That is still my wish.) The
main effect of this change is to delete a few definitions from basic.tex and
other macro packages.

Here are the names of the integer parameters:
\tolerance	(formerly \chpar1=)	badness tolerance after hyphenation
\pretolerance	(formerly \chpar15=)	badness tolerance before hyphenation
\hyphenpenalty	(formerly \chpar2=)	hyphenation penalty
\finalhyphendemerits			penultimate line hyphenation demerits
		(formerly half of \chpar3=)
\doublehyphendemerits			double-hyphen demerits
		(formerly half of \chpar3=)
\widowpenalty	(formerly \chpar4=)	widow line penalty
\brokenpenalty	(formerly \chpar5=)	broken line at page end penalty
\binoppenalty	(formerly \chpar6=)	math binary op break penalty
\relpenalty	(formerly \chpar7=)	math relation break penalty
\predisplaypenalty  (formerly \chpar9=)	penalty for breaking before a display
\mag		(formerly \chpar12=)	1000 x magnification ratio
\adjdemerits	(formerly \chpar13=)	adjacent incompatibility demerits
\looseness	(formerly \chpar14=)	change in paragraph length
\uchyph		(formerly \chpar16=)	uppercase hyphenation
\exhyphenpenalty  (formerly \chpar17=)	explicit hyphenation penalty
\day		(new)			initialized to current day of month
\month		(new)			initialized to current month of year
\year		(new)			initialized to current year
\time		(new)			initialized to minutes since midnight
\interlinepenalty    (new)		see below
\postdisplaypenalty  (new)		see below
\displaywidowpenalty (new)		see below

The former \chpar10 (dump window) is no longer needed, since TeX82 has better
ways to display token lists.  The former \chpar11 (\radsign) goes away in favor
of the new \radical primitive.  The former \chpar18, \chpar19, \chpar20, once
"reserved for extensions", are gone too, since it is now best for a TeX extender
to give names to whatever new parameters are needed. Similarly, \x is gone.
* The \tracing parameter disappears, and its various components each have their
own names.  They are as follows:

\showboxbreadth (nodes per level when a box is being exhibited)
\showboxdepth (maximum level shown when a box is being exhibited)
\pausing (if nonzero, lines from a file are displayed as they appear)
\tracingonline (if nonzero, diagnostic info goes to terminal as well as to file)
\tracingmacros (if nonzero, shows macros as they are being expanded)
\tracingstats (if nonzero, shows memory usage when TeX has recorded it)
\tracingoutput (if nonzero, shows boxes when they are shipped out)
\tracinglostchars (if nonzero, shows chars dropped because they aren't in font)
\hfuzz (a dimen parameter; hboxes are reported if more overfull than this)
\vfuzz (a dimen parameter; vboxes are reported if more overfull than this)
\hbadness (under/overfull hboxes exceeding this (integer) badness are reported)
\vbadness (under/overfull vboxes exceeding this (integer) badness are reported)

* The new \tolerance and \pretolerance are devalued by a factor of 100 from the
old \jpar and \jjpar.  In other words, the default is now \tolerance 200 and
\pretolerance 200, so that it is a true "badness tolerance", i.e., the badness
should not exceed 200. Any value of 8128 or more is equivalent to an infinite
value, in which glue can stretch arbitrarily far.

* Furthermore, \codeval and \parval are eliminated. In their place is a much
more powerful operator called \the. For example, what used to be "\codeval5" is
now "\the\chcode5"; what used to be "\parval2" is now "\the\hyphenpenalty". You
can even say "\the\hsize" to get the current \hsize as a dimension,
"\the\baselineskip" to get the current \baselineskip as a glue value; and
you can say things like "\vbox to \the\baselineskip", which makes a vbox
whose height is the normal amount of baselineskip (exclusive of stretching
and shrinking). When expanding macros, you can say "\the\font" to get the
current font identifier (e.g., "\f"), as well as "\the\f" to get the
corresponding font name, as well as "\the\output" and "\the\everypar".
In the latter cases, macros inside the current output or everypar routines
are not further expanded.

* Things like \count and \dimen and \skip should not appear except in the
context of numbers, dimens, and glue; for example, you shouldn't say "\dimen 5"
or even "\count 5" in the midst of a paragraph. But if you say "\the\count5",
the paragraph will get the (signed decimal) value of the counter; if you say
"\the\dimen10" you will get text like "3.14159pt"; and "\the\skip12" yields text
like "-5.00000pt plus 3.40001fil". And "\number\count5" gives the conventions of
TeX80, i.e., lower case roman numerals if the counter is negative.  In \xdef and
\write, TeX82 will expand occurrences of \the and \number, using the current
values, but \count and \dimen and \skip will not be expanded. This gives you a
little more control over what gets expanded.

* \thebox is changed to \lastbox (avoids confusion with \the).

* \minusthe is like \the, but gives the negative value.
* An array of 256 dimension values is introduced, called \dimen0 to \dimen255.
Furthermore `3.5dm8' is a dimension equal to 3.5 times \dimen8.  These join
\count0 to \count255 and \skip0 to \skip255, so we now have 256 of each basic
quantity (instead of 10, as in TeX80).  This applies to \count, \dimen, \skip,
\write, and \box.  PLAIN.TEX contains macros for allocating a new \count or
\dimen, etc.

* mu-glue and ordinary glue are unmixable, since the rules are so much cleaner
and clearer this way: There are 256 \muskip registers, which take glue whose
units are mu (instead of pt, etc.).  It is now illegal to do things like
\hskip\the\thinmskip or \mskip\the\baselineskip.

* Operations on \count, \dimen, \skip, and \muskip are extended, so that we now
have a complete set:

	\setcount <digit> [=] <number>
	\advcount <digit> by <number>
	\multcount <digit> by <number>
	\divcount <digit> by <number>

	\setdimen <digit> [=] <dimen>
	\advdimen <digit> by <dimen>
	\multdimen <digit> by <number>
	\divdimen <digit> by <number>

	\setskip <digit> [=] <glue>
	\advskip <digit> by <glue>
	\multskip <digit> by <number>
	\divskip <digit> by <number>

	\setmuskip <digit> [=] <mathglue>
	\advmuskip <digit> by <mathglue>
	\multmuskip <digit> by <number>
	\divmuskip <digit> by <number>

Note that \specskip has changed its name to \setskip. The division operations
truncate towards zero.  The = sign in \setcount, \setdimen, \setskip, \chcode,
\font, \openout, \let, and in other similar things, is now optional.
Furthermore an equals sign is optionally allowed now after \tolerance,
\hsize, \baselineskip, \setbox (the new name for \save), etc.
* If you put \global in front of \def or \let or \chcode or \tolerance or
\baselineskip or even \parshape or \hangindent or a font identifier, the
definition will now be global. Otherwise the definition is local (except
for \gdef and \xdef).  This is a change in the case of dimension
parameters:  \varunit, \parindent, \lineskiplimit, \mathsurround,
\maxdepth, \topbaseline, \hsize, \vsize; you should put \global in front
of these to get the former behavior. You probably wanted the former
behavior only when changing \hsize or \vsize in an \output routine.

* \edef is a local \xdef. Both \edef and \xdef can now take arguments like \def
and \gdef.

* Two other modifiers can be placed in front of \def, \gdef, \edef and \xdef:
	\long means that the arguments to the macro are allowed to contain
		\par tokens; formerly this was always allowed, but now it
		is permitted only for "long" macros. Otherwise TeX will now
		stop when it sees \par going into an argument, presuming
		that a right brace was forgotten. This detects one of the
		most frequent errors made by TeX users, before it propagates
		to overflow the memory.
	\outer means that the macro being defined is not allowed to appear
		subsequently either in an argument or in the right-hand side
		of a definition or write text, or in the preamble of an
		alignment. In other words, the macro should appear only at
		"quiet" times. This is another way to catch missing braces
		before too much damage is done. It used to be applied at the
		end of 	every page, but most TeX users don't use a page-oriented
		editor like E; therefore TeX82 does not treat file pages as
		an integral part of its control structure.

* New primitive \mathchardef will save lots of memory if you have lots of
control sequences for math things like \alpha, etc.:
	\mathchardef\cs[=]<fifteen bit number>
will have the effect of
	\def\cs{\mathchar <fifteen bit number> }
and it will use none of TeX's memsize. (By contrast,
	\def\cs{\mathchar "1234 }
takes up 8 words of memory!)

* \groupbegin and \groupend provide an alternate way to enter and leave groups
for locally defined values. A \groupbegin will not match a }, nor will { match
\groupend; the former gives the message "Missing \groupend inserted" when the }
occurs, and the latter inserts a "missing" }. Note that you can introduce
\groupbegin in one macro and \groupend in another.
* If you say \message{text}, the terminal will display " text" immediately.  For
example, the new version of PLAIN.TEX contains message statements so that
when INITEX inputs the file your screen looks something like this:
	(PLAIN.TEX preloading macros, fonts, codes, hyphenation)
instead of "(plain.tex 1 2 3 4 5 6)". (Note that you don't say \input basic any
more, and PLAIN.TEX is already preloaded when you run TeX, as explained below.)

Here's an example macro that displays names of sections when you get to them in
a paper you are TeXing:
  \outer\def\section#1{\vfill\eject\message{#1}\ctrline{\bf#1}}

* You can also say \errmessage{text}, which causes a TeX error message like
	! text.

* \mathcode<n> replaces \chcode <n+128>.

* A new chcode value 14 denotes a character that is better for comments than the
present code 5. A character of code 14 denotes end of the current line (i.e.,
ignore the remainder of that line), without inserting a blank space, and without
considering that line to be all blank.  Thus, if % is assigned type 14, you can
have lines that are completely comments by starting them with %, without having
this line come out as \par; and you can also end a line with % without having a
blank space inserted there.

* Another new chcode value, 15, denotes an invalid character. When such a
character is input, TeX82 issues an error message.

* Here's something that was NOT put into TeX82: It wouldn't be hard to make TeX
understand \escape to mean 0, \opengroup to mean 1, ..., \active to mean 13,
\comment to mean 14, and \invalid to mean 15; then you could say, e.g.,
\chcode'14=\active in the example above. But it seems wrong to make \chcode too
easy, since that will only encourage more people to fiddle with the \chcode
table. Let's leave this a black art, to be resorted to only with reluctance in
times of emergency.

* You CAN do certain things now in horizontal mode, e.g., \vfill; TeX82 will
silently insert the \par you forgot.

* \discretionary{#1}{#2}{#3} makes discretionary characters other than hyphens.
It means the text should either contain #3 without a break, or else it should
contain #1, then a break, and then #2. For example, \- is equivalent to
\discretionary{-}{}{}. The parameters #1, #2, and #3 may not contain anything
but letters and otherchars (not spaces or penalties, etc.); they need not all be
in the same font, and TeX will insert ligatures and kerns within them if
necessary. For example, the correct way to specify the hyphenation of
"difficult" is "di\discretionary{f-}{fi}{ffi}\-cult". In German, the correct way
to specify hyphenation of "backen" is "ba\discretionary{k-}{k}{ck}en";
presumably if we were doing a lot of these we would define a \ck macro so that
one could type "ba\ck en" or "ba{\ck}en".  The third part of a \discretionary
must be empty, in math mode.
* The internal character set used by TeX82 is the same regardless of the
external character set. There is no longer a difference like "\chcode'176" for
right brace that applies only at SAIL! Right braces and underlines and tildes
and notequals and a few others have been a source of problems that have now gone
away. Furthermore there is now a way to input an ascii control character to any
version of TeX82 by typing, e.g., ↑↑A.

TeX82 assumes that all of the standard ascii characters, shown in positions 040
through 176 below, are available; these characters are always converted to their
standard ascii codes. For example, a TeX user who types A is asking for
character 65 of the current font, even though the A might have entered the
computer in EBCDIC or some other code. Non-standard-ascii characters might also
be readable on some implementations of TeX. In such cases they should have the
significance stated below, for best results; and all characters that cannot be
converted to a compatible TeX code should be converted to 177.
	ascii	TeX	description	chcode	mathcode (when TeX starts)
	000	↑↑@	null		ignore	bin401
	001	↑↑A	downarrow	submark	rel443
	002	↑↑B	alpha		other	ord213
	003	↑↑C	beta		other	ord214
	004	↑↑D	and		other	bin536
	005	↑↑E	not		other	ord472
	006	↑↑F	epsilon		other	ord217
	007	↑↑G	pi		other	ord231
	010	↑↑H	backspace,lambda ignore	ord225
	011	↑↑I	tab,gamma	space	ord215
	012	↑↑J	linefeed,delta	ignore	ord216
	013	↑↑K	uparrow		supmark	rel442
	014	↑↑L	formfeed,+/-	endline	bin406
	015	↑↑M	carriage-return	endline	bin410
	016	↑↑N	infinity	other	ord461
	017	↑↑O	partial		other	ord245
	020	↑↑P	subset		other	rel432
	021	↑↑Q	superset	other	rel433
	022	↑↑R	intersection	other	bin534
	023	↑↑S	union		other	bin533
	024	↑↑T	for-all		other	ord470
	025	↑↑U	there-exists	other	ord471
	026	↑↑V	circle-times	other	bin412
	027	↑↑W	doublearrow	other	rel444
	030	↑↑X	leftarrow	other	rel440
	031	↑↑Y	rightarrow	other	rel441
	032	↑↑Z	notequal	other	rel434
	033	↑↑[	escape,diamond	action	bin567
	034	↑↑\	less-or-equal	other	rel424
	035	↑↑]	greater-or-equal other	rel425
	036	↑↑↑	equivalence	other	rel421
	037	↑↑_	or		other	bin537
	040		space		space	ord464
	041	!	exclamation	other	close041
	042	"	double-quote	other	close042
	043	#	hashmark	param	ord561
	044	$	dollar-sign	math	ord577
	045	%	percent-sign	comment	ord045
	046	&	ampersand	align	ord046
	047	'	apostrophe	other	close047
	050	(	left-parenthesis other	open050
	051	)	right-parenthesis other close051
	052	*	asterisk	other	ord052
	053	+	plus-sign	other	bin053
	054	,	comma		other	punct054
	055	-	hyphen,minus-sign other	bin400
	056	.	period		other	ord056
	057	/	slash		other	ord057
	060	0	zero		other	ord060
	   . . .
	071	9	nine		other	ord071
	072	:	colon		other	rel072
	073	;	semicolon	other	punct073
	074	<	less-than-sign	other	rel074
	075	=	equal-sign	other	rel075
	076	>	greater-than-sign other	rel076
	077	?	question-mark	other	close077
	100	@	at-sign		other	ord574
	101	A	uppercase-A	letter	ord301
	   . . .
	132	Z	uppercase-Z	letter	ord332
	133	[	left-bracket	other	open133
	134	\	backslash	control	bin404
	135	]	right-bracket	other	close135
	136	↑	caret		supmark	ord017
	137	_	underline	submark	ord465
	140	`	reverse-apostrophe other open140
	141	a	lowercase-a	letter	ord341
	   . . .
	172	z	lowercase-z	letter	ord372
	173	{	left-brace	open	open546
	174	|	vertical-line	other	ord552
	175	}	right-brace	close	close547
	176	~	tilde		other	rel430
	177	↑↑?	invalid		invalid	ord573

As before, the mathcodes (which replace Appendix F8 of the old TeX manual) are
relevant only when the chcode is letter or other. (See below for the new chcode
values.) Two possibilities are given for codes 010, 011, 012, 014, 033, 055; at
most one of these should be chosen, and if both are present on some system
keyboards the other should probably be disallowed for TeX input (mapped into
177). However, since a user can change any chcode and any math chcode, strict
conformity with these interpretations isn't absolutely necessary. To convert a
file into a format that all TeXes can read, one should change null into ↑↑@,
downarrow into ↑↑A, and so on. If a character set contains uparrow but not caret
(e.g., the SAIL system falls into this category), the uparrow should be
considered an ascii caret; code 013 will be used only if both uparrow and caret
are present, as they are at MIT.  Incidentally, this internal coding scheme is
based on a scheme used at MIT, since the MIT code is faithful to ascii while
allowing additional visible characters that are extremely convenient.

An appearance of ↑↑A is equivalent to an appearance of ascii code 001, if the
current chcode of ↑ is supmark. In particular, if somebody in a foreign country
with more than 26 letters in the local alphabet wants to make \chcode ↑↑A =
letter, then control sequences like \a↑↑A↑↑Ab (a four letter word) are
permissible.

TeX82 puts 015 (ascii carriage-return) at the end of each line, except for the
lines that are inserted with "i" after error messages. If the final character of
the line is currently chcoded to be an escape character (e.g., if you end an
error-insertion with \, or if you do \chcode'15=0), the result is control-null,
which is an undefined control sequence unless you define \↑↑@.

Of course, users are expected to type \ne instead of ↑↑Z if their system's
character set doesn't contain a not-equal sign; TeX82 recognizes ↑↑Z as ascii
032 primarily to make it possible for straightforward translation of TeX files
from one system so that they will work on another.

Some files contain ascii 014 (form-feed) characters as page marks. Such
characters are ordinarily treated like carriage-returns, since the initial
chcode for 014 is carret. In order to get TeX82 to do the error checking at the
end of a page, as the old TeX did, you can say
	\chcode'14=13 \outer\def↑↑L{\par}
* Note that the backslash character is now predefined as an escape character
when TeX82 begins. The old idea about letting the user's first nonblank
character be the escape has been abandoned. Furthermore TEXPRE has been replaced
by a version of TeX called INITEX that allows an entire macro package to be
preloaded; this macro package can define its own chcodes and mathcodes. The
normal version of TeX already has "PLAIN.TEX" preloaded; the normal version of
AmSTeX already has the AmSTeX macros and fonts preloaded.

The new rule for starting TeX is this:  When you're running INITEX or a version
of TeX that has a preloaded format, you can request format file f.fmt by typing
`&f' after the ** prompt. When you're running VIRTEX, format file plain.fmt will
be loaded unless you type `&somenonplainformatname' after the ** prompt.  Thus,
for example, the following ways of starting TeX are equivalent at SAIL and
similar sites:
	tex paper
			r tex;paper
					r tex
					** paper
							r tex
							**\input paper
(The asterisks here are TeX's prompt character. On TOPS-20 what used to be
"@tex paper/", indicating batch mode, is now "@tex \batchmode\input paper".)

* There are four new primitives \batchmode, \nonstopmode, \scrollmode,
\errorstopmode that represent increasing amounts of interaction. \batchmode and
\nonstopmode will never stop for any reason; \batchmode omits printing anything
on the terminal (but the .err file gets everything, as usual).  These nonstop
options are intended for overnight batch processing.  \scrollmode doesn't stop
for error messages, but it does stop if files can't be found, or if \pausing is
nonzero. \errorstopmode is the default. If you aren't in \errorstopmode, your
.err file will contain "help" messages for all of your errors. These modes are
global (they don't revert at end of group).  They can be set in a format file;
thus you can have a format that implies batch processing. But you could override
that, e.g. by running "batchtex \errorstopmode \input paper". (This example
assumes that batchtex is a program representing "virtex batch", i.e., a virgin
TeX with batch.fmt loaded and then the core image saved.)

* \dump will save TeX's current memory contents.  \dump is essentially like \end
(it's the last thing you do with INITEX), and you don't specify a file name. If
your input was named foo, your output file will be named foo.fmt. Such files are
now called format files.  This is allowed in INITEX only, and only at very quiet
times (i.e., at group level 0 in vertical mode with nothing on the current page,
etc.).  The file name will be printed later when these memory contents are
loaded in a production version of TeX; for example, if you say "\dump" on
March 1, 1982, the TeX that uses the dumped file might begin with the line
	This is TeX, Version 1 (format=plain 82.3.1)

* Actually the program name TeX now stands only for versions of TeX82 that have
PLAIN.TEX preloaded. Other preloaded versions (e.g. AMSTEX) will usually exist
too. If your operating system does not allow a program to start with its memory
preloaded, you will have to call a "virgin TeX" program VIRTEX that first wants
to see the name of a format-dump file (e.g., PLAIN or AMSTEX).  In this case a
typical calling sequence might be "@virtex &amstex paper".   If no format
is given, "&plain" is assumed. If your operating system is nice enough to
allow preloaded programs, a typical way to create the program TeX would be
to say "@virtex &plain" followed by something like "control-C" and "save tex".

* \indent takes you from vertical mode to horizontal mode and indents the
paragraph; this can be used if the first item in the paragraph is in an
\hbox or \vbox. You can also use \indent in horizontal mode to stand for
"\hbox to\the\parindent{}".

* If you end the parameter part of a definition with an additional #, the
argument-matching process will terminate on the next left brace.  For example,
in
	\def\chop to #1#{\chopp{#1}}
the call "\chop to 2in{x}" will expand to "\chopp{2in}{x}". The definition
	\def\mac#{why}
will subsequently issue an error message if "\mac" is not followed by "{".

* The "texinfo" that is given with each font (see Appendix F of the METAFONT
manual) can now be changed by a TeX user program; there was no way to do this
before except by making a new TFM file. Say
	\texinfo <font><parameternumber>=<dimen>
For example, \texinfo\a3=4pt sets the stretch component of spacing to 4pt in
font \a. (Parameter 1, the "slant", is unitless but you should give its value in
units of points.) You can use this feature to adjust math-mode positioning of
subscripts, etc., by changing the parameters in mathsy and mathex fonts. Note
that texinfo is global, it does not get reset at the end of a group. You can
also say \the\texinfo<font><parameternumber>.

* New dimension parameters \hfuzz and \vfuzz specify the tolerance for printing
a diagnostic message about overfull boxes. If the box is overfull by this amount
or less, no message is printed. Default is ".1pt", which was the old TeX
standard. If you say \hfuzz 8000pt, you probably won't see any overfull boxes.
If you say \hfuzz 0pt, you will see all of them, including a few that you didn't
know about last year.

* Another new dimension parameter \overfullrule specifies the width of a rule
that is added at the right end of overfull hboxes. This rule has the height and
depth of the box. If \overfullrule is zero or negative, or if the amount of
overfullness does not exceed \hfuzz, no rule will appear.  Default is 5pt, which
gives a big black mark to help you spot overfull boxes.

* The "overfull box" warning messages will be given in a new form that simply
gives the characters in the box; for example,
	Overfull \hbox, 3.3326 points too wide (in paragraph of lines 210--216):
	\f This is the text of a line that was over-full for some rea-son.
Discretionary hyphens are shown as real hyphens, so that you can see what
hyphenation TeX was trying.  The error-transcript file gets both this message
and an old-style description of the overfull box in detailed diagnostic dump
format.

* Another new control sequence, \relax, does nothing at all. Thus, if you want
to disable the action of a control sequence, you can \let it be \relax.
* Up to 256 fonts may be used, and each font may contain up to 256 characters.
(Characters numbered 128 to 255 can be accessed either via ligatures or
charlists or with the \char command.)

* \leftskip and \rightskip specify glue to be placed at the left and right of
each line of a paragraph. This provides better ways to do ragged right setting,
and it makes changes to \hsize less necessary.

* \lastskip gives the value of the previous item in the current horizontal or
vertical list, if that item was glue, otherwise it yields the value 0pt. Thus,
to get the effect of \unskip in vertical mode, say
"\penalty100000\vskip\minusthe\lastskip".
You can also do things like "\ifdim\the\lastskip > 5pt{...}\else{...}", so that
one macro can make decisions about spacing based on what has gone before.  (This
finally solves the long-standing problem about spacing after theorems that end
with a displayed equation.) Note that \the\lastskip is permitted, except in
\write statements.

* \sqrt signs in TeX82 are positioned differently in their boxes:  The baseline
now comes exactly at the bottom of the place where the vinculum (i.e., the rule
over the operand of \sqrt) is to be joined.  This means that no rounding errors
will be possible and perfect alignment will be obtained at all resolutions.
Pre-82 versions of TeX will still work (subject to rounding) if the height of
the box is the thickness of the rule.

* The error transcript files are no longer called "errors.tmp". Your output file
and transcript file will be "paper.dvi" and "paper.log" if your first line of
TeX input specifies \input paper. The default name "texput" is used whenever no
other appropriate name has occurred before TeX reads line two of its input. (TeX
can't wait any longer, since line one has to be put into the transcript file,
and the transcript file has to have a name before it gets information.)

* \hyphenation{word list} can be used to override TeX's hyphenation algorithm;
for example, to specify hyphenation of the words "hyphenation" and
"exceptions" one can write
	\hyphenation{hy-phen-a-tion ex-cep-tions}
A new hyphenation algorithm devised by Frank Liang is used in TeX82; this one
extends much more readily to other languages. Words containing ligatures can now
be hyphenated automatically, even difficult words like "difficult".

* If two fonts are specified with the same name and point size, only one will be
loaded.

* \sfcode <char number> is the spacefactor code for that character, times 1000.
For example, the spacefactor code for period and question mark is normally 3000,
for comma 1250, for right parenthesis 0 (meaning do not change the space
factor), and for most characters it is 1000. In TeX82, you also say
"\spacefactor 1234" instead of "\spacefactor 1.234".
* TeX82 has a new "help" facility available on error messages. If you type "h"
after an error, you will (usually) get further explanation of what the error
means, together with suggestions about how to proceed.

* \↑↑\ and \↑↑] have gone away, to the delight of people who don't have nice
ways to type ascii control characters. Instead, \nonscript in math mode precedes
a space of any other type, making that space zero in subscript styles. Thus, the
conditional thin space is now "\nonscript\mskip\the\thinmuskip", and conditional
negative thin space is "\nonscript\mskip\minusthe\thinmuskip".

* \thinmuskip, \medmuskip, \thickmuskip are now definable like other glue
parameters such as \baselineskip. The units should be in mu. For example, one of
the defaults is \thickmuskip 5mu plus 5mu.  The old "mathspace" parameter in
symbol fonts (see METAFONT manual p99) is no longer used.

* There's a new way to get up to 4096 more math symbols in all three sizes, by
defining font families 0 to 15. For example, suppose that fonts \A, \D, and \F
are Fraktur alphabets in 10pt, 7pt, and 5pt sizes. Then you can say
	\textfont 5=\A \scriptfont 5=\D \scriptscriptfont 5=\F
which is something like the code "\mathrm adf" in the old basic.tex. Now if you
say "\fam5" in math mode, you get characters from font \A, \D, or \F, depending
on the size. For example, "{\fam5 B↓b}" would give Fraktur B in 10pt with a
subscript Fraktur b in 7pt. The rule is that a family specification overrides
the math chcode for symbols of type letter and otherchar.

Note that we can now say \def\rm{\fam0\f} together with
	\textfont 0=\f \scriptfont 0=\g \scriptscriptfont 0=\h
and then it's possible to say, e.g., "\def\max{\mathop{\rm max}}" instead
of resorting to "{\char`m \char`a \char`x}" in order to achieve size-switching.
This extension also makes ligatures and kerning available in math mode.

* The use of certain families is predefined.  Family 2 specifies the `mathsy'
fonts used for symbols; family 3 specifies the `mathex' fonts used for large
delimiters; and the other 14 families can be used in any desired fashion.
Instead of `\mathsy uxz' one now says
	\textfont 2=\u \scriptfont 2=\x \scriptscriptfont 2=\z
and these assignments are local (they go away at the end of a containing group).
You must have \textfont2, \scriptfont2, \scriptscriptfont2, \textfont3,
\scriptfont3, and \scriptscriptfont3 defined before using math mode, since the
parameters of these fonts contain the values TeX needs for math spacing.
(\scriptfont3 and \scriptscriptfont3 are now supposed to be math extension
fonts, as well as \textfont3, because TeX82 will use smaller extension-type
features in script and scriptscript styles; for example, the default rule
thickness in a subscript is a parameter to \scriptfont3, while TeX80 had only
one extension font for all three sizes.)

\fam1 is implicit whenever TeX enters math mode.
* The \mathcode now has a 15-bit number as its value. The first three bits
specify ord, op, bin, rel, open, close, punct, and var (where var is like ord
but it subsitutes the "current" family for the stated one).  The other twelve
bits specify a "math character", with four bits for the family and eight for the
character. For example, character '100 in family 5 is '2500; of course, this
reads a little better in hexadecimal:  character "40 in family 5 is "540.

You can say \mathchar followed by such a 15-bit code, to get the equivalent of
typing a character with that math code. Thus, one can now say, e.g.,
	\def\cdot{\mathchar '10001 }
instead of \def\cdot{\mathop{\char1}} as formerly. The control sequence \char
is no longer allowed to take values greater than 255.

* When math mode is entered, the current family is set to 1. Thus, family number
1 is generally for math italic fonts.

* "\comb" is changed to "\atopwithdelims". There's also "\overwithdelims" and
"\abovewithdelims".
For example, {1\overwithdelims[]2} is sort of like \left[1\over2\right].

* The \radsign parameter goes away. Instead, one says \radical followed by a
delimiter code. Delimiter codes may be used also after the control sequence
`\delimiter' in connection with \left and \right and \atopwithdelims.

A delimiter code is a somewhat esoteric 24-bit number. The first twelve bits
specify a `small' character, and the last twelve bits specify a `large' one.
When TeX chooses a delimiter, it searches in the following way until finding the
first one large enough: First it looks at the `small' character in the current
size of the family, then (if the current size isn't text size) it looks at the
small character in the next larger size, and so on until coming to text size. If
a suitable delimiter has still not been found, the same search is carried out
starting at the `large' character. If any of the characters looked at is part of
a "charlist", the list is searched before moving on. If the small or large
character is zero, it is ignored; thus, you can't use character 0 in family 0 as
a delimiter.

For example, \sqrt is equivalent to \radical '11601560 in the Computer Modern
fonts; the 1160 specifies \fam2\char'160, and the 1560 specifies \fam3\char'160.
Since '160="70, we can also write this as \radical "270370.

If \delimiter <delimiter code> appears in a formula in some place not
controlled by \left or \right or \atopwithdelims, it is actually a 27-bit
code. The least significant 12 bits are ignored, and the leading 15 bits
are used like a \mathchar (thus, they specify a category as well as a
family and character). The reason is that one can now say, e.g.,
	\def\lfloor{\delimiter '411421404 }
so that one can say both \left\lfloor and simply \lfloor.

A \delcode is also given for letters; \left and \right and \delimitedatop will
use this code as a delimiter code if they are followed by an "otherchar".  For
example, the Computer Modern fonts use \delcode`(='501400, assuming that family
0 contains the ordinary roman alphabets.  Initially, \delcode is negative for
all characters; this denotes an invalid delimiter.

All this bit hackery is, of course, unfriendly looking, but the goal is to make
it possible for macro packages to define the friendly codes without taking up
much memory space inside of TeX. All of the TeX control sequences that used to
be predefined for Computer Modern are now unbundled so that arbitrary encodings
can be used. One of the embarrassing limitations of TeX80 was that it could
handle delimiters only between '142 and '153 in the symbols font, and it
insisted that these `small' delimiters had corresponding `large' ones in
positions '004--'015 of the mathex font!
* The spacing in math formulas is unbundled too; there are three parameters
\thinmuskip, \medmuskip, \thickmuskip to specify the spacing in formulas like
$x\log x$, $x+x$, and $x=x$, respectively. One defines these using "mu" units,
e.g. `\medmuskip = 4mu plus 2mu'. The control sequences `\,' `\>' and `\;' in
math mode yield spaces of these three varieties. (These are now defined in
PLAIN, not TeX primitives. The previous meaning of \> is now rendered
`\nonscript\>'.)

The following example shows a feature that is NOT allowed:
$a \save1\hbox to 18mu{}
	\ifdim 1wd1=10pt{\gdef\x{\over b} \else{\def\x{}}
	\x$
Whoever wrote this was trying to be clever and discover whether TeX was
in \textstyle. But the program is self-contradictory, because if the "a" is in
10pt text style the formula changes itself to $a\over b$ where the "a" is in
script style, so the formula changes itself to $a$ where the "a" is in text
style, so... Constructions like this show why TeX does not allow variable
dimensions like mu except in very restricted ways like \mskip.

* New dimension parameters
	\scriptspace  {this amount is placed at the right of all subscripts
			and superscripts; TeX80 used about .45pt always}
	\nulldelimiterspace {this amount is used before and after all fractions
			defined by \atop, \over, \above, and for all
			"." delimiters}
	\delimiterlimit {see the next parameter}

* New integer parameter \delimiterfactor. When TeX computes the size of \left
and \right delimiters, it computes delta1=twice the maximum distance of the
enclosed formula from the "axis". (The axis is where the fraction line would
go.)  Let delta2=delta1*(delimiterfactor/1000) and delta3=delta1-delimiterlimit.
The delimiters will be as small as possible provided that their height+depth
exceeds both delta2 and delta3. (TeX80 took delimiterfactor=900 and
delimiterlimit=1ex; TeX82 lets the user twiddle with these magic numbers.)

* When unscripted letters occur in math mode, they now are adjusted for
ligatures and kerns. This means, for example, that "df" will be spaced better,
once appropriate kerning information has been added to the math italic fonts.
The (new) rules for spacing are this: If there's no kerning specified, add the
italic correction to every symbol in math mode; otherwise use the kern (without
the italic correction). However, a subscript is moved left by the amount of
italic correction. In formulas like $P↓2↑2$, the 2's will no longer be directly
above each other (the sub-2 will be to the left of the sup-2 by the amount of
the italic correction).  TeX80 put them above each other, thereby following a
long-standing convention (cf. Oxford book by Chaundy et al.), but this usually
turned out to be undesirable, so people started to write $P↑2↓{\!2}$ all the
time.  If anybody really wants it the old way, they can get it by
$\hbox{$P$}↑2↓2$.
* The maximum penalty has been raised from 1000 to 10000; thus, "\penalty 1000"
no longer absolutely prohibits a break, but "\penalty 10000" does.  This number
10000 is being used elsewhere in TeX82 also: For example, \tolerance or
\pretolerance of 10000 is equivalent to saying "use all possible line breaks,
regardless of how much stretching is necessary."  A \penalty of -10000 (or less)
is equivalent to the old \eject in vertical mode, and to the old \linebreak in
horizontal mode. These are no longer primitives of TeX82.

* The \pagebreak feature is also eliminated, in favor of a much more general
feature. You can say \vadjust{vertical list} in the midst of any paragraph, and
whatever is in the specified vertical list will be placed immediately following
the box for its line when the paragraph has been made. For example, \pagebreak
is now written \vadjust{\penalty-10000}. You can use this feature to do things
like insert extra space between lines of a paragraph, something like \noalign
does for alignments.

* \ifabsent n tests if \box n is not present. All boxes are initially absent;
a null box (made e.g. by \hbox{}) is empty but present. A box becomes absent
after it is used.

* \topbaseline has changed to \topskip; thus glue is allowed at the top of a
page. This makes it easy to "bottom justify", for example.  (The old \topskip
and \botskip are no longer used.)

* \vtop is now allowed in any mode, just like \vbox.

* \special{keyword arg} is a general extension feature. The keywords are system
dependent, but TeX copies "keyword arg" into the DVI file so that any device
driver that knows your keywords will do the right thing with them. Users should
get together if they want to standardize on various keywords. Examples:
\special{halftone fig22} could mean "insert a halftone from file fig22, with its
reference point at the current reference point"; \special{leftend 2} and a later
appearance of \special{rightend 2} could mean "draw a straight line from
the left reference point to the right one" (the "2" is an identifier to
distinguish this line from another one); \special{message Foo} could mean
"display `Foo' on the console of the printing device"; and so on.
Semantically, \special acts like a box of height, width, and depth zero,
as far as TeX is concerned; the argument in braces is sent to the DVI file
where it is associated with the current reference point.
The length of "keyword arg" must be at most 255 characters.

* \ragged is no longer implemented, since \rightskip does ragged right setting
so much better.

* \case <number> {case0}{case1}{case2}\else{remaining cases} illustrates a new
way to choose between more than two alternatives without too many nested
brackets.
* Major changes have been made to the \output conventions, so that it will be
easier to produce balanced columns and various other things.  The old ideas of
\topinsert, \botinsert, \topsep, \botsep are eliminated!

In their place one says \insert n, where n is a box number, e.g. "\insert 250".
Different numbers correspond to different classes of insertions; for example,
one might want to have figures as well as footnotes inserted at the bottom of
pages, and TeX80 used to use the same treatment for both. Under the new
conventions, all \insert 250's that go on a page will appear in \box 250 when
the \output routine starts.

The old idea of "\page" is gone too, and the meaning of \output has changed, so
read carefully: The contents of an accumulated page, exclusive of inserts, is
placed into \box 255, so that the output routine can place this material
together in whatever way it wants. (\insert 255 is not legal.)  A new TeX
primitive called \shipout, followed by a box specification (e.g.,
"\shipout\vbox{\box255\box250}") is what actually produces output.  Note that
\ifabsent 250 can be used to test whether any \insert 250's have been gathered
for a page.  The \shipout command can be used anywhere, not just in \output.
Incidentally, \shipout prevents the old anomalies about the values of \counts
being different from what people thought they would be when \writing
table-of-contents or index data to a file. The default \output routine in TeX82,
if none is specified, is "\output{\shipout\box255}".

Another new primitive, \vsplit, is handy for multi-column input.  The command
"\setbox 2=\vsplit 250 to 100pt" will, for example, make \box2 a box whose
height is 100pt, by extracting 100pt worth of material out of \box250. The depth
of \box2 will be at most \splitmaxdepth (using the rules that TeX used for
\page); \vsplit extracts the optimum initial segment of \box250, in the sense
that badness+penalty is minimized and the segment is as long as possible subject
to this condition. After \vsplit has acted, \box250 will contain the residual,
as if a page break had occurred; thus, glue and penalties will be eliminated
following the break, and the first box or rule (if any) will then be preceded by
sufficient glue to position its top baseline (based on \splittopskip).

The main idea of \vsplit is to make it easier for an \output routine to produce
multiple column format. For example, if \box255 is 300pt high, one can get
triple columns by
	\setbox 1=\vsplit 255 to 100pt
	\setbox 2=\vsplit 255 to 100pt
	\setbox 3=\vsplit 255 to 100pt
(Actually it is safer to use slightly less than 100pt here, but the exact
measurements depend on the top baseline and other things.) The remaining
material in box 255, if any, can be put back onto the following page, as we will
see momentarily; and boxes 1,2,3 can be positioned as desired before they are
shipped out.

The \vsplit routine also looks at \marks in the contained box. It sets
\splitfirstmark and \splitbotmark to the topmost and bottommost contained marks;
otherwise it sets these to null strings.

Of course you can't apply \vsplit to a box that was constructed as an \hbox.
There is no \hsplit.

The old \output routine defined a sequence of items in restricted vertical mode,
with the meaning that this sequence would be vboxed and shipped out.  The new
\output routine defines a sequence of items in restricted vertical mode, with
the meaning that this sequence will be placed in front of whatever TeX has
accumulated for the following page, including whatever caused a break on the
current page. Thus, if you write "\output{\unbox255}" you are in serious
danger of getting in an infinite loop.

Consider, for example, what happens if a page break occurs at some glue.  If the
output routine leaves some of the tail end of the material from \box 255 in its
vertical list, this material will fit perfectly before the glue that caused the
previous break, since the glue following a break is not eliminated when \box255
was made; glue is simply discarded when it appears at the top of the new page
that is started after \output finishes.

Consider also what happens if a page break occurs at "\penalty-10000". If you
understand what has just been said, you will see that this would cause repeated
looping if the \output routine places something back for the next page, and this
might be a problem. Therefore TeX will change the penalty to +10000 at a break,
and it also sets \outputpenalty to the value of the penalty that actually caused
the break. (\outputpenalty is set to 10000 if it was glue that caused the
break.) Thus, you can restore the effect of a penalty by putting
"\penalty\outputpenalty" at the end of your output routine.

But how does TeX figure out what to put in box 255 and how much of the
insertions to put into other boxes, before it calls on your output routine?  The
rules are slightly complicated, but they have been devised to handle a wide
variety of situations, including situations where some inserts span several
columns. It should be possible to do things like put footnotes in two columns
beneath single-column text, or to put single-column footnotes beneath
double-column text, etc. Here's how:  We associate \skip n, \dimen n, and \count
n and \box n with \insert n.  (a) \count n gives a magnification ratio of
\insert n with respect to ordinary text. For example, if two-column footnotes go
with one-column text, and if footnotes are inserted with \insert250, then
\count250 should be 500. If single-column footnotes or page-wide figures are
being inserted with double-column text, the magnification ratio should be 2000.
(b) \dimen n gives a maximum length of inserts for box n; subsequent inserts
will be carried over to a following page. (c) \skip n gives a correction term
when there is at least one insertion for box n.

The total length of inserts is figured as the sum, over all n such that \insert
n appears on a page, of the \skip n plus (\count n over 1000) times the total
natural height plus depth of all \insert n's, including the original contents of
box n before any inserts were made. The badness of a page is computed from the
amount of the text on the page plus the total length of inserts, and TeX breaks
the page so that badness is minimized. The \vsize is the total of text plus
insertions; if you are using double-column text, your \vsize should be about
twice the actual page height.

If this isn't complicated enough, there is also a rule for splitting inserts, so
that long footnotes can be broken between pages and so that large figures can be
carried over to subsequent pages. Here's the idea: When we are deciding whether
to perform an \insert n or not, we first look to see if previous \insert n's
have all been completed without splitting.  If not, this one is carried over to
the next page. If so, this one is put on the current page, if it does not cause
the page to overflow and if it does not cause the maximum (\dimen n) to be
exceeded. In the latter cases, the insertion is \vsplit to the maximum size that
would not cause such overflow. For example, suppose we get to an insertion at
magnification 500 when \vsize minus the current amount of text and the previous
total amount of insertions leaves only 50pt of vertical space left. Suppose the
insertion takes 150pt of vertical space, so that it would take 75pt after
scaling; and suppose that 150pt would not exceed the maximum total size of
insertions for this box. Then we essentially \vsplit the insertion list to 100pt
(this will scale down to 50pt), after which the actual length of the insertion
will be computed as its natural height plus depth (which might be different from
100pt). The remaining part of the insertion will be corrected for top baseline,
etc., as in an ordinary \vsplit, if this broken insertion is actually chosen.
But TeX will only use broken insertions if they lead to the minimum badness for
the resulting page.

If several \insert n's appear on the same page they are concatenated together
with no baselineskip correction between them. So you should use struts to
produce the correct positioning.

Insertions are treated the same whether they appear in horizontal or vertical
mode. A "floating" insertion turns out to be a special case of a broken
insertion, whose first component is a null box.

TeX doesn't put waiting insertions into \box255, it leaves them on the list for
the subsequent page. If insertions appear in a box that is being vsplit, they
are ignored.

It's too bad that these rules came out so complicated, but in simple cases the
output routines will now be quite simple, and the manual will have enough
examples to make things clear (I hope). Nothing simpler than this seems to
provide the other features that people have been demanding, and the total amount
of programming for the TeX82 page builder is not much more than there was in
TeX80.
* In vertical mode, you can say \prevdepth=3pt to make TeX82 act as if the
previous box had a depth of 3pt, when computing the glue between boxes to
achieve the baselineskip. If you set \prevdepth to a value less than or equal
to -1000pt, the baselineskip calculation will not be made. (This is the
case at the beginning of a vbox, or just following an hrule.)

* You can say \the\spacefactor and \the\prevdepth, if you are in horizontal or
vertical mode, respectively.

* New diagnostic features: "\showbox 10" will display the current contents of
\box10 on the terminal. There's also \showthe as in "\showthe\count 5" and
"\showthe\baselineskip". Also "\show\cs" to give a symbolic display of the
current meaning of the control sequence \cs.  The present \ddt is deleted, and
\showlists exhibits the current activities the way \ddt used to.  If you are
trying to diagnose some mysterious behavior, you can say, for example,
"\showthe\texinfo\f 40" and you will get an error message like "\font \f has 7
texinfo parameters" (if it has fewer than 40). Incidentally, if your error
message was "\font\g has 7 texinfo parameters", you would know that fonts \f and
\g are being treated identically. (TeX loads only one copy of a font that you
mention twice. If you want to load two distinct copies, so that you can diddle
their parameters independently, you can try something like this:
"\font \f=cmr10 at 10pt  \font \g=cmr10 at 10.00002pt".)

* If you use a \skip parameter in the context of a dimension, the natural width
is used. For example, \setdimen 5=\the\baselineskip. If you use a \dimen
parameter in the context of an integer, the conversion is in units of sp (scaled
points, 2↑{16} of a pt). For example, \setcount 10= 1truept would set \count 10
equal to 65536000 divided by \mag.

* When you use \advskip, infinite glue wipes out finite glue.
For example, "\setskip 2=5pt plus 2pt minus 1fill
	      \advskip 2 by 3pt plus 1fil minus 1fil"
is equivalent to "\setskip 2=8pt plus 1fil minus 1fill".

* \linepenalty is yet another parameter to control line breaking. TeX82 adds
this to the badness before squaring to get demerits. (Previously, I had
\linepenalty=1 always; by setting it a bit higher, like maybe 7 or 8, you tend
to get paragraphs that are set tighter when a line can be saved.  I don't think
it's a good idea to make \linepenalty real large, and it would be foolhardy but
weird to make \linepenalty = -10000; this apparently would minimize the number
of lines but maximize the badness!)

* Macro parameters are now delimited by strings instead of single items.
For example, \def\a#1ab{...} followed by \a acaab will set #1 to "aca";
\def\a b#12#{foo#1} followed by \a bbar2baz2{8} will expand to foobar2baz{8};
the latter followed by \a 2... will give an error message (\a not followed
by b). You get the error message only if there's a string before the
first parameter.
* Characters chcoded 13 are not equivalent to one-letter control sequences. They
act like control sequences (e.g., you can use them after \def and \let), but &
and \& will be distinct.

* \everypar{...} inserts its argument into TeX's scanner at the moment TeX has
changed from vertical to horizontal mode. The paragraph indentation will already
appear in the paragraph, unless of course the transition to horizontal mode was
due to \noindent.

* Spanned and omitted columns in alignments: If an entry in an alignment is
`\omit', the preamble text for the column is omitted in this row.  The control
sequence \span can be used in place of a tab mark, and the result is that the
surrounding entries are combined together.  You can use \omit only as the first
item of a column; "&\omit\cr" is equivalent to "\cr", as formerly, but now
you are allowed to omit columns other than the first.

Example: \tabskip 1em plus 1em
	 \halign to 20em{\ctr{#}&\rt{#}&\lft{#}\cr
				AAA&B&C\cr
				DDDDDDD\span\omit&EE\cr
				\omit\span FFFFF\span\omit\cr
				G&\omit H\span III\cr}
where all the letters are 1em wide, say. Let wij be the maximum width of the
entries that span columns i thru j.  The first line "AAA&B&C" implies that
w11≥3, w22≥1, w33≥1. The second line says that w12≥7 and w33≥2, and so on;
we find that w11=3, w12=7, w13=5, w22=1, w23=4, w33=2. Column widths are
now assigned from left to right, as follows:
	c1=w11
	c2=max(w22,w12-t1-c1)
	c3=max(w33,w23-t2-c2,w13-t1-c1-t2-c2)
where ti is the natural width of tabskip between columns i and i+1.  In this
case t1=t2=1, so c1=3, c2=3, c3=2. This means the natural width of the lines
will be 1+3+1+3+1+2+1=12 ems, so the glue will be stretching to make up the
additional 8ems; each unit of stretch is doubled. When columns are spanned,
however, TeX justifies the material into a box having the appropriate width for
the tabskip glue that was omitted; for example, an entry that spans columns 1
and 2 will be justified to width c1+t1+sf*s1+c2, where s1 is the stretchability
between columns 1 and 2, and sf=2 in this case since the glue is being doubly
stretched. (If the tabskip glue shirinks, we would of course use shrinkability
instead; in this case spanned columns might actually get smaller than their
natural size.)

The result of the above example, taking account of which parts of the preamble
are omitted by the \omit operations, is therefore
	|   AAA     B   C    |
	|    DDDDDDD    EE   | 	(columns 1 and 2 spanned and centered)
	|            FFFFF   |  (columns 1 to 3 spanned, right justified)
	|    G   HIII        |  (columns 2 and 3 spanned, left justified)
Restriction: A single entry can span at most 256 columns.

* Spaces are ignored after tab marks in alignments.
* \hbox par is eliminated! Instead of "\hbox par 100pt{...}", one now says
"\vbox{\hsize 100pt ...}" and the effect is almost the same. The only difference
is that the paragraph or paragraphs in the \vbox will have both their hanging
indentation and looseness (and perhaps also their baselineskip) specified inside
the vbox; this is, of course, more logical than the old rule.

Restricted vertical mode is no longer restricted; it's called "internal vertical
mode". It differs from vertical mode only in not going through the page builder,
and in allowing \unskip as well as \vskip with infinite shrinkability.

* To have the paragrapher work on inserted text (e.g., in footnotes), one writes
"\insert 250{\hsize 200pt ...}", perhaps using the \interlinepenalty parameter
that adds to the penalty between lines of a paragraph (whether in inserts or
not).

* Three new dimension parameters \displaywidth, \displayindent, and
\predisplaysize are assigned values at the beginning of every displayed formula:
(1) displaywidth is the length of the line that will contain the formula before
it is centered; this is usually equal to \hsize, except when hanging indentation
or \parshape are being employed in a paragraph.  (2) \displayindent is the
amount by which that line is indented.  (3) \predisplaysize is the amount of
copy on the line preceding the displayed formula; this is what is used to decide
between \dispskip or \dispaskip.  If the display immediately follows \noindent
or another display, \predisplaysize will be -(2↑{30}-1) sp (the smallest legal
dimension in TeX). Otherwise, if the position of the last box on the previous
line is affected by glue stretching or shrinking, \predisplaysize is set to
+(2↑{30}-1) sp. Otherwise \predisplaysize is set to the natural width that the
preceding line would have if all glue were removed at its right end, plus the
amount of indentation of that line, plus 2ems.

* \setcount, \setdimen, \setskip, \setbox, \output, and \everypar are local
definitions unless specified global. Likewise the results of \advcount, etc.

* The { } in \output still defines grouping (it would be too dangerous to
leave it out, since \output occurs asynchronously), but the { } in \everypar
does not. Grouping is now independent of \if tests, as explained later.

* All of TeX's primitives now mean the same thing in all modes. The control
sequences that were exceptions to this rule have been dealt with as follows:
\  (control space) now means a text space, even in math mode.
\quad is no longer a primitive (PLAIN defines it as \hskip 1em).
\! is no longer a primitive (PLAIN defines it for math mode only)
and \ignorespace is a new primitive that gobbles spaces.
\- is always a discretionary hyphen; its previously advertised mathmode
function has been taken over by the new \nonscript primitive.
* \accent is allowed only in horizontal mode; \mathaccent only in math mode.
The latter takes a 15-bit math code, the former an 8-bit character code; it's
like the difference between \char and \mathchar.

The \mathaccent primitive will make use of a charlist of characters to choose
the first accent of a list whose successor is either nonexistent or wider than
the formula being accented. Thus, you can have a list of longer and longer
tildes or hats, etc.

* Spacing in math is slightly different: Boxes and {...} and \left...\right
subformulas are given type Inner, so there are eight types instead of seven.
(Previously Inner was treated like Ord.) The spacing matrix entries are set so
that x followed by Inner is the same as x followed by Open; Inner followed by x
is the same as Close followed by x.  This removes the anomaly that "\left("
didn't act like "\mathopen(".  (The new fonts have more space outside of
parentheses, especially outside of the large parentheses.)

* "\vcenter to 100pt" is now allowed in math mode, if anybody wants it.

* "\ifdimen" is renamed "\ifdim" and there's also "\ifnum" replacing "\ifpos".
Examples: \ifnum\count1>5{...}\else{...}; \ifdim\dimen3<1.5wd2{...}\else{...}.

* "\ifinner" is true if the current mode is internal-vertical,
restricted-horizontal, or non-display-math. Combining this with \ifvmode,
\ifhmode, \ifmmode makes it possible to determine exactly what mode you are in.

* New primitive \kern allows you to specify unbreakable space (without
stretching or shrinking). Thus, "\kern -1pt" is something like
"\penalty10000\hskip-1pt". There's also "\mkern 3mu" in math mode.  You are
allowed to use \kern but not \hskip in \discretionary lists.  A \kern in a word
does not upset the hyphenation algorithm.  You can use \kern in vertical as well
as horizontal lists.  It is legal to break at a kern if it is immediately
followed by glue or leaders, provided that it is not preceded by glue, kern, or
penalty.

* New debugging facility \tracingcommands, if nonzero, gives a symbolic
indication of what commands are being obeyed by TeX's main control routine.

\dpenalty disappears in favor of \postdisplaypenalty
\displaywidowpenalty is a new thing, it replaces \widowpenalty
	just before displays. (In TeX80 this penalty was zero.)

* New integer parameter \maxdeadcycles gives an upper limit on how many
consecutive invocations of \output do not cause at least one \shipout.  This is
intended to catch unintended loops. Default is 25.

* If \tracingstats > 0, you get to see how close you came to TeX's current
table capacities, in a list of statistics printed at the end of your run.
If \tracingstats > 1, you also get to see the current memory usage
every time you do a \shipout.
If \tracingstats > 2, you also get a huge amount of inscrutable printout
about what the line-breaking algorithm thinks it is doing.
But \tracingstats is ignored unless TeX has been compiled in a "slow version"
that actually maintains these statistics.
(The SAIL version of TeX is currently "slow" in this way.)

* \string replaces the next token by its text, with all characters regarded
as type otherchar (except that a space will be of type spacer; it's possible
but not easy to get a space here). For example, \string\abc results in four
characters \, a, b, c. This expansion occurs just as for \number, i.e., when
\string occurs in \xdefs or in horizontal or math mode. (The most common
use of \string is to follow it with a macro parameter.) Caution:
If characters in a control sequence name are nonstandard in ascii, they
will be converted differently at different installations.
* Here is an extension to the language intended to placate people who
have objected to the fact that \write (and \openout and \closeout) only
cause action at the time of the next \shipout. Some applications
call for immediate output, hence a new feature: \immediate followed by
\openout or \write or \closeout causes the output action to take place
without delay. For example, \immediate\write{x} is equivalent to
\shipout\vbox{\write{x}} except that the latter also puts an empty
page into the DVI file.

* New parameter \boxmaxdepth affects \vbox: If the depth of the box
would exceed \boxmaxdepth according to the normal rules, the box contents
are shifted up so that the depth is exactly \boxmaxdepth, before setting
the glue. The same applies to \vtop, before adjusting its depth.
The default setting is \boxmaxdepth='7777777777sp (the maximum dimension).

* You can now use \let with non-control-sequences after the = sign.
For example, \let\zero=0 makes the control sequence \zero behave something
like the digit zero; but if you want to make the constant 100 by saying
"1\zero 0" you still have to \def\zero{0}. Thus, this new extension isn't
a big breakthrough, but it does save a bit of space and time inside TeX.
(Incidentally, after \let\zero=0, \zero will not expand to 0 in xdefs.)

* Popular demand wins again: You can now say \csname <string>\endcsname
to manufacture a control sequence name. For example, \csname foo \endcsname
is essentially identical to a control sequence named "\foo " (note that
the space is part of that name!) and \csname foo\endcsname is like "\foo"
and, after \def\zero{0}\def\test{zero}, it follows that
\csname\csname\test\endcsname\endcsname is like "\0". The conversion
from token list to control sequence occurs as if \csname were a
macro being expanded. If the control sequence hasn't been defined before,
it will behave as if it were "\relax".

* In the preamble to \halign or \valign, the primitive \span would normally
make no sense. But it causes TeX to expand the following token, instead of
just copying it, before inserting that token in the preamble.
(Previously this was possible only with a dirty "\tabskip" trick, since
TeX expands whatever follows "\tabskip 0pt" looking for "plus".)
* Syntactic conditionals!
For years people have been asking for TeX to treat conditionals in its
"mouth" rather than in its "stomach", and I have been fending them
off. But starting with Version 0.8 of TeX82, \if tests are made at the
time of macro expansion rather than as part of the semantic processing in
horizontal or vertical or math mode.
(Note: Previous pages of this listing still use the old \if notation;
I didn't want to bother to update it.)

Instead of writing \if...{..a..}\else{..b..} the new syntex is
	\if... ..a..\else..b..\fi
(with \else optional if ..b.. is empty).  Instead of writing
\case...{..a0..}{..a1..}{..a2..}\else{..b..} the new syntax is
	\ifcase... ..a0..\or..a1..\or..a2..\else..b..\fi
(and again the \else is optional).

Whenever TeX is reading material in a mode where macros are now expanded,
it will process conditionals somewhat as though they were macros. Namely,
\if... results in evaluating the condition and skipping code if the condition
isn't true (skipping to the next \else that isn't enclosed by \if..\fi
brackets); \else, \or, and \fi switch in the appropriate way between
reading and not reading text.

Braces need not be properly nested inside the conditionals, nor do \if...\fi's
need to be properly nested in the replacement texts of macros. Thus,
for example, \xdef\lbrace{\if TT{\else}\fi} \xdef\rbrace{\if TF{\else}\fi}
define macros that expand to singleton left and right braces.
(Having these two types of nesting independent of each other has proved
to be important in many existing macro processors. Caveat implementor.)

People who try things like \expandafter\else\if... should be shot.
(Unless it turns out that this is useful?)

* Another token-list parameter \tokens is definable like \everypar.
Its only use is for things like \the\tokens (which, in an \xdef,
emits the current value of \tokens without further macro expansion).

* Replace "chcode" by "catcode" everywhere above.

* Another test, \ifcat, is like \if but it tests the catcodes of the
characters, not their ascii codes.
* \everymath{...} inserts its tokens into TeX's scanner just when
non-display math mode has been entered. \everydisplay{...} does likewise, but
for displays. For example, \everymath{\fam0} sets up family 0 instead of
family 1 as the default for letters; you can also use \everymath
to redefine active characters that you want to behave differently
in math mode.

* \futurelet\a followed by tokens b and c has the effect of "\let\a= c"
followed by tokens b and c. You can use this to look ahead at the
next token after a macro; it's for hackers. (I put it in because it
is easy and because it might allow me to solve some problem next year.)