path: root/info/digests/tex-implementors/030

Date:	  8 April 91				Message No:	030

To:	  TeX implementors and distributors

From:	  Barbara Beeton

Subject:  bug reports and replies from DEK


This message contains only one thing: a file of bug reports that was
sent to Knuth, annotated with his replies (typed in by me from his
handwritten notes).  Please note that "sections" of this file are
separated by form feeds (^L); I hope they don't give anyone's machine
indigestion.


########################################################################

a quick summary of what is included here:
  TeX.WEB 3.1
    Brian {Hamilton Kelly}, spurious error history with |show_whatever|
    Wayne Sullivan, non-printing characters in font identifiers
    Eberhard Mattes, dvi files with more than 65535 pages
    Barbara Beeton, file found by \input not found by \openin
  PLAIN.MF
    Harald B\"ogeholz/Eberhard Mattes, problem when aspect_ratio <> 1
  TeXbook
    Robert Hunt, possible bugs in \ifcat test, page 377
    Robert Hunt, difficulty in appendix G, rule 14
  GFtype/GF format
    Peter Breitenlohner, GFtoPK finds an error where GFtype does not;
         inconsistencies in GF format and programs reading/writing GF files
  MF.WEB, PK doc
    Edgar Fuss, precision incompatibilities
  PLtoTF/VPtoVF
    Wayne Sullivan/Chris Thompson, a PLtoTF/VPtoVF bug
  TANGLE
    Peter Breitenlohner, token memory statistic error while digesting input


		*****	 TeX.WEB 3.1	*****

Date: Thu, 13 SEP 90 10:10:00 BST
Reply-to: Brian {Hamilton Kelly} <TeX@rmcs.cranfield.ac.uk>
From: TEX@rmcs.cranfield.ac.uk
Subject: Feature that's really a bug in TeX v3.0 (and earlier)

The routine |show_whatever| at section 1293 is invoked for any variety
of \show primitive.  Eventually, it calls |error| to complete any
necessary interaction with the user (if desired).  However, the latter
routine:
(1) always incrments |error_count|, and will crash with a fatal error if
    there have been 100 ``errors''
(2) always sets |history := error_message_issued|

The code of |show_whatever| pre-decrements |error_count|, but {\em
only\/} if not interacting with the user: therefore it's feasible that a
file could contain 100+ \show commands, and if the user were patient
enough, would eventually get the fatal error exit.

But more importantly, more sophisticated versions of TeX which pass back
the value of |history| to the operating system will be misled into
thinking than a genuine error has occurred, and perhaps the operating
system will take some (in)appropriate action.

I can see two solutions:
(1) ensure that |error_count| is always predecremented in
|show_whatever|, and also preserve the value of |history| before calling
|error|, and restore it afterwards: however, this sort of external
interference with global variables is surely to be deprecated
(2) make |error| take a parameter, which indicates whether the routine
is being called because of a genuine error, or merely to provide an
opportunity for interaction: this would more correctly modularize the
funtionality, but involve changes to many dozens of other calls to
|error| within the program.

I discovered this when attempting to understand why my VMS
implementation appeared to swallow the remainder of a DCL command
procedure with which I was attempting to automate the TRIP test: I'd
mistakenly thought that TeX was keeping open the |termin| which was in
this case taking its input from the body of the command procedure.  In
the case of the TRIP test, of course, the exit status was genuinely
correct in reporting that |error_message_issued|, and the cure was to
tell DCL not to take error exit from the command procedure; however, in
course of investigating this with a smaller TeX source, taken again from
a command procedure, I realized that any \show command also caused the
exit to occur with error status: and after all, the help for any \show
is ``This isn't an error, I'm only showing something...''!

                               Brian {Hamilton Kelly}
-------

[ coment by DEK:
  Granted, the logic isn't ideal.
  but I was aware of these deficiencies and I don't consider them serious
  enough to warrant any change at this late date.  [METAFONT is by the way
  exactly the same w.r.t. _showing_, except that it clears error_count
  after statements rather than paragraphs.]
  The patient user might indeed have a problem with more than 100 show
  commands, but there is a way to work around it i a pinch by forcing TeX
  to think that a paragraph has ended.  Or you can type
  `i\error\errorstopmode\show ...' then `q'.
]

Date: Wed, 05 Dec 90 15:40:28 GMT
From: "Wayne G. Sullivan" <WSULIVAN%IRLEARN@pucc.PRINCETON.EDU>
Subject: TeX Bug

BUG REPORT TeX WEB 3.1 :

In sections 234, 267, 579 and 1322 of TeX 3.1, there appears
     print_ecs(font_id_text(...))
          [esc  -- correction by DEK ]       
or the equivalent.   Since font identifiers can have non-printable
characters, one of the slow print procedures should be used instead.

                Wayne Sullivan, December 1990
-------
Date: Sat, 08 Dec 90 20:47:06 GMT
From: Chris Thompson <CET1@phoenix.cambridge.ac.uk>
Subject: Wayne Sullivan's new TeX bug

FYI here are the comments I sent him, and his reply.

 ----------------------------------------------------------------------

You are absolutely right. Simplest test I have found that shows the
effect (from section 267) is

    \expandafter \font \csname abc^^80def\endcsname = cmr10
    \setbox0=\hbox{XYZ} \tracingonline=1 \showbox0

One gets a space where one should get ^^80.
         [ ^^^^^ or system crash depending on system  -- DEK ]

I believe that you have identified all the cases that need amending.
All other calls of |print_esc| are either on constant strings made
up of printable ASCII, or one can prove that the variable argument
is a string so made up. (There are three such cases: in |print_cs|
and |sprint_cs| the calls |print_esc(p-single_base)| must refer to
the printable ASCII representations at the bottom of the string pool;
and in |print_write_whatsit| the argument |s| is an explicit constant
string in each of its calls.) Given that, there would probably be a lot
to be said for inventing a new |print_font_id| routine: one might use
this to cover up some of the early convoluted forward references at
the same time.

Chris Thompson
JANET:    cet1@uk.ac.cam.phx
Internet: cet1%phx.cam.ac.uk@nsfnet-relay.ac.uk

 ----------------------------------------------------------------------

Knuth actually catches the important case: \the\font . However, at some
                                              [^^^^^  because it fetches
                             the `permanent' font identifier which is
                             treated like any control sequence  -- DEK ]
stage it would be worth cleaning up the others. I think sprint_cs does
the trick, but a more ornate resolution of the problem may be the
ultimate outcome.   Would you forward a copy of your note to BNB? I sent
her the same message I sent you, but it could possibly have been missed.
                              Thanks,
                                Wayne
-------

[ comment by DEK:
  Since print_esc is never `inner loop' I decided that it
  should call slow_print.  This saves a bit of code in \S262-3 too.
  By the way, the reward for TeX bugs has reached its max now!  So it
  is not keeping up with inflation and the falling $/\lb ratio, sorry.
                              $326.78  \S63
  Now slow_print is used _only_ in print_esc.
]

Date: Fri, 28 Dec 90 15:44:53 +0100
From: Eberhard Mattes <mattes@azu.informatik.uni-stuttgart.de>
Subject: How to report TeX bugs

Dear Ms. Beeton,

I've found a very very small but old (maybe there has never been a TeX
version without this bug!) bug in TeX. What's the correct/official way
of reporting it?

Yours,
       Eberhard Mattes   (mattes@azu.informatik.uni-stuttgart.de)
-------
(replied that if he sent it to me, i would ask chris thompson to vet it,
and then forward the results to dek)
-------
Date: Fri, 28 Dec 90 16:25:01 +0100
From: Eberhard Mattes <mattes@azu.informatik.uni-stuttgart.de>
Subject: How to report TeX bugs

Here's the bug (thanks for forwarding it to Chris Thompson and DEK!):

total_pages isn't checked for overflow: A dvi file may contain only up
to 65535 pages. If one feeds a document with more than 65535 pages
into TeX, either the dvi file is written incorrectly (total number of pages
modulo 65536 in the postamble) or TeX bombs at

  dvi_out(total_pages div 256); dvi_out(total_pages mod 256);@/

as dvi_out expects a eight_bits number. Maybe it's a limitation rather
than a bug. But it should be documented.

Yours,
       Eberhard Mattes (mattes@azu.informatik.uni-stuttgart.de)
-------
Date: Sat, 05 Jan 91 16:28:17 GMT
From: Chris Thompson <CET1@phoenix.cambridge.ac.uk>
To: bbeeton <BNB@math.ams.com>
Cc: Eberhard Mattes <mattes@azu.informatik.uni-stuttgart.de>
Subject: DVI files with more than 65535 pages

It is certainly true that TeX will generate an invalid DVI file (if
it doesn't get a range check first---that will usually depend on
compilation options) if more than 65535 \shipout commands are obeyed.
What I don't know is what Don's attitude is likely to be towards this
as a bug report. For example, he lets himself off a number of painful
hooks in the comments in section 104:

  The present implementation of TeX does not check for overflow when
  dimensions are added or subtracted. This could be done by inserting a
  few dozen tests of the form `if x>='10000000000 then report_overflow'
  but the chance of overflow is so remote that such tests do not seem
  worthwhile.
             [ I was not thinking of PicTeX when I wrote that, but still
               I am of the opinion that 2^{15} points aer plenty big  -- DEK ]

(Actually, it is quite easy to provoke such overflows deliberately, and
I have occasionally had reports of it happening accidentally.) On the
other hand, he did add quite a lot of code (#383) in TeX 3.0 to prevent
demerits overflowing, which, despite what Frank Mittelbach says, I have
some difficulty believing is a practical problem. (EM's bug is surely
                             [ ^^^^^^^^^ believe me it is!  -- DEK ]
*not* a practical problem, is it? It is certainly 100 times as many
pages as I have ever TeX'ed at once!)

Eberhard Mattes sent me a message saying, inter alia
> here are four suggestions for fixing the bug (in decreasing order of
> (my) preference):
>    - ignore (and document) the bug
>    - change dvi format: total_pages = 65535 means 65535 or more
>    - change dvi format: total_pages = 0 and final_bop <> -1 means
                                      [ ^ This alone should suffice to
               distinguish it in TeX output, but not other routine  -- DEK ]
>                         65535 or more.
>    - error message

I don't think changing the DVI specification even in this very minor
respect is on at this stage: instead, one has to take the line that a
valid DVI file cannot contain more than 65535 pages. The `correct' fix
would, I think, be a fatal error (of the |overflow| type) on attempting
to produce the 65536th page.

Anyway, I think you should you send it to Don and see what he says about
it. If he does make a TeX change as a result, I suppose I will feel an
obligation to add a similar check to my program for merging several DVI
file into one: it certainly doesn't at the moment!

Chris Thompson
Cambridge University Computing Service
JANET:    cet1@uk.ac.cam.phx
Internet: cet1%phx.cam.ac.uk@nsfnet-relay.ac.uk
-------

[ comment by DEK:
  I decided that this isn't serious enough to warrant a
  change to DVI format [and we do want to save trees];
  nor to give an overflow error [which would make the TeXbook incorrect];
  so I'll adapt Everhard's first preference, to ignore and document
  the bug (but to put total pages need 65536 in DVI file just to
  avoid bombing out when TeX is someday put onto a chip).
  Surely max_push will _not_ exceed 65535 ...
                              $2.56  TeX \S642
]

Problem found at AMS by Barbara Beeton, December 1990

I have a multi-file LaTeX job where i need to keep the input files
segregated, so I simply made up a suitable search path for tex$inputs
and worked in another directory.  The main file input to LaTeX contains
several statements \include{file.name} using the technique recommended
in the LaTeX manual, rather than equivalent \input file.name statements.
The indicated files are not found, but if \include is replaced by \input,
the retrieval is successful.  This is the behavior on a VAX/VMS system.

LaTeX uses \openin to first check for the presence of \include'd
files, to permit absent files to be bypassed rather than hanging the
job.

I next copied the files involved to our DEC-20, and tried again.
Under TOPS-20, all files were found by the \include/\openin .  This
led me to believe that the problem is system-dependent.  I discussed
the matter with Dave Kellerman, whose VMS implementation we are using.
He verified that the handling of files with \input and \openin is
indeed quite different, and that it seems to be consistent with the
(slender) documentation in the .WEB file.  However, he said that some
changes had been made to \input and it wasn't clear to him whether
DEK intended to leave \openin unchanged or it was an oversight.  Given
the way that LaTeX is using \openin (which seems quite reasonable), it
may be an oversight.  This conclusion is given more weight by the fact
that David Fuchs' TOPS-20 implementation used the search path for both
\input and \openin even though it wasn't in the original .WEB.

I sent an inquiry to Chris Thompson, and here is his analysis.  This
is an area that seems to be too easily interpreted differently by
different implementors, and an authoritative ruling would be useful.
-------
Date: Sun, 06 Jan 91 00:36:53 GMT
From: Chris Thompson <CET1@phoenix.cambridge.ac.uk>
Subject: \input vs \include (aka \input vs \openin)

The underlying difference in unmodified TeX is that \input (implemented
in |start_input|) tries first the file name as given, and then (if the
specified area field was empty, since fix #312) using |TeX_area| as the
area. \openin (implemented in |open_or_close_in|), on the other hand,
just tries the name as originally specified, leaving \ifeof<n> true
if it can't open it. How implementors map this behaviour in terms of
current directories, paths, or whatever their operating systems provide,
is no doubt a rich and varied feast. (For the record, in my MVS
implementation, the "try the default area" behaviour is mapped onto
"if a simple name is specified, try it first as a DDname, and then as
a member name in a standard PDS concatenation", and I do maintain the
difference between \input and \openin.)

I am fairly sure that the rationale for the difference is that \openin
is paired with \openout (which you clearly wouldn't want to have the
area defaulting behaviour: you want it to create the file where you
said, if it doesn't exist already). It seems very likely that any `real'
use of \openin, i.e. with the intention of using \read, would be reading
a file created by \openout and \write in the current or a previous run
of TeX.

LaTeX, on the other hand, uses \openin in a much more sneaky way, to
obtain a `softly-failing \input' function. The relevant macro, \@input,
is used not only to read \include'd files, but also when reading back
the .aux file(s), .toc file, etc. If the \openin fails to open the file
(as tested by \ifeof), LateX outputs a warning message; otherwise it
does a \closein and uses \input in the sure and certain knowledge that
it will find the file. (This is an expensive way of doing things, in
operating systems that make opening a file a non-trivial activity.
It has occurred to me that there is scope for some ingenuity in TeX
implementations in not actually doing an operating system close on
a stream that is \closein'ed while in the state |just_open|, but
keeping it around in the hope that a \input for the same file name
will follow shortly! I haven't actually tried to code this, though.)

With the difference between \input and \openin as in unmodified TeX,
then it is certainly true that if \openin succeeds, so will \input
(unless another process is deleting files in parallel, of course), but
the converse is not true. File names passed to LaTeX's \@input have
the semantics associated with \openin, not that associated with \input.

I hadn't realised that as many implementations as you say have modified
the file name semantics of \openin to be the same as \input. Maybe this
is becoming a de facto standard. I am rather doubtful whether it is
really the correct thing to do: one should not forget \openin's original
purpose, as well as LaTeX's peculiarities. It is certainly possible to
argue that it is a desirable change even in that context, though. I
suppose it is too large a modification to consider at this stage the
addition of a conditional-\input primitive to TeX itself.

Chris Thompson
-------

[ comment by DEK:
  My current opinion is:
    If the operating system allows users to define a ``custom''
      search path at run time, then both \input and \openin
      should be able to use it, although I would hope that
      people don't use \openin for `system' files but only for files
      they tend to control themselves.
    If the operating system is like WAITS (on which I developed TeX),
      where there's no decent way to provide a clue to TeX at runtime
      about a nonstandard search path, then I would provide access
      to the main system macro files (like plain.tex and
      webmac.tex) only for \input not \openin; I would
      use the same strategy to search user's personal files
      for both \input and \openin.
  I have found it _very_ useful with UNIX to put `..' on the
  standard search path.  Then I can create a subdirectory
  called say _pages_ and cd to pages, on which I can
  run TeX/MF with some temporary changes to input fies and I won't
  clobber any of the master files or the parent directory.
  My applications of this idea would fail if \openin didn't
  also look at .. directory when unable to find an . directory.

		*****	  PLAIN.MF	*****

Date: Mon, 7 Jan 91 11:38:54 +0100
From: Eberhard Mattes <mattes@azu.informatik.uni-stuttgart.de>
To: CET1@phoenix.cambridge.ac.uk
Cc: BNB@math.ams.com
Subject: Now a real bug

Dear Ms. Beeton, dear Mr Thompson,

now I am reporting a real bug:

plain.mf:

 ----------------------------------------------------------------------
def drawdot expr z = if unknown currentpen_path: def_pen_path_ fi
 addto_currentpicture contour
  currentpen_path shifted z.t_ withpen penspeck enddef;
 ----------------------------------------------------------------------

does not work correctly with aspect_ratio <> 1. Here's a fix:

 ----------------------------------------------------------------------
def drawdot expr z = if unknown currentpen_path: def_pen_path_ fi
 addto_currentpicture contour
  currentpen_path shifted (z.t_) withpen penspeck enddef;
 ----------------------------------------------------------------------

BTW, local.mf of emTeX has been containing this fix for over a year,
and nobody reported this bug. It was found by Harald B\"ogeholz (HWB),
                                            [ ^^^^^^^^^^^^^^^^^^^^^^^
                                                  $2.56 MFbook p271  -- DEK ]
a friend of mine.

Yours,
    Eberhard Mattes (mattes@azu.informatik.uni-stuttgart.de)
-------
Date: Mon, 07 Jan 91 15:29:24 GMT
From: Chris Thompson <CET1@phoenix.cambridge.ac.uk>
To: Eberhard Mattes <mattes@azu.informatik.uni-stuttgart.dbp.de>
Cc: Barbara Beeton <BNB@math.ams.com>
Subject: Re: [Now a real bug]

Dear Eberhard, & Barbara,

Yes, I agree with that: the bracketing should be

  currentpen_path shifted (z transformed currenttransform)

not, as effectively at present,

  (currentpen_path shifted z) transformed currenttransform

Will you pass that on the Don Knuth, Barbara?

I have sometimes wondered why mode_setup doesn't make t_ into a dummy
symbolic token when currenttransform (after yscaling by aspect_ratio)
is the identity, in the same way as o_ is redefined. This would save
some time in common cases; currently you have to use 'notransforms'
yourself to achieve it. I suppose it would be an incompatible change
by now, though, in that METAFONT users may be relying on being able
                      [ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ yes indeed indeed -- DEK ]
to alter currenttransform at any time.

Chris Thompson
JANET:    cet1@uk.ac.cam.phx
Internet: cet1%phx.cam.ac.uk@nsfnet-relay.ac.uk
-------

		*****	   TeXbook	*****

Date: Mon, 21 Jan 91 17:07:05 GMT
From: Chris Thompson <CET1@phoenix.cambridge.ac.uk>
Subject: TeXbook p.377 bug(?)

  I have had what apparently looks like an error in the TeXbook
reported to me by a user here, Robert Hunt <reh10@phx.cam.ac.uk> of
the Department of Applied Maths and Theoretical Physics. He claims
that in the macro in the footnote on page 377 of the TeXbook, the
\ifcat test

  \ifcat\noexpand#1\noexpand~\explicitfalse % active funny space

can never give the result true, so that the line is unnecessary (and
misleading). I think he is right, but this area of TeX is distinctly
unsusceptible to intuition! I intend to advertise on UKTeX and TeXhax
a challenge to find arguments to \stest that will make this \ifcat
deliver true; meanwhile could you please record that Robert has a
first claim if this does indeed turn out to be a TeXbook bug?

Chris Thompson
-------

[ comment by DEK:
  $2.56  He is correct that the test never succeeds
  but he is wrong to assume that it is unnecessary ...
  The present code has a bug (it might say that an active
  implicit funny char is explicit) as in
  \uccode`=`*\catcode`*=\active
    \uppercase{\uppercase{\let*=}}
    \newtoks\t \t={*123}
  which leads to the
    \global\explicittrue  branch.
  TeXbook 20\th printing will have corrected code, namely
    \long\def\ssss#1#2\stest{\funnytrue
      {\uccode`#1=`~
      \uppercase{\ifcat\noexpand#1}%
      \noexpand~\global\explicitfalse
      \else\escapechar ... \fi\fi}}
  which makes the points I had hoped to make in the original
  footnote and preserves the idex entries to {\it 377}.
  I suspect this will be the all-time champ for obscure bug in
  The TeXbook ...  But that's the whole point of this footnote.
]

Date: Sun, 27 Jan 91 22:08:12 GMT
From: Chris Thompson <CET1@phoenix.cambridge.ac.uk>
Subject: Another TeXbook error (or obscurity)

Robert Hunt (he of the p.377 example) has come up with another (sigh)
possible TeXbook error. I think his case is pretty good on this one.
He asks why          [ p 444 ]

    \setbox0=\hbox{$V.$}\showbox0

produces

    > \box0=
    \hbox(6.83331+0.0)x9.16667
    .\teni V
    .\kern2.22223            (this is the italic correction for \teni V)
    .\kern-1.66667           (this is the kern between the V and the .)
         [ ^^^^^^^ oops, maybe I should have made this more negative ...
                         too late  -- DEK ]
    .\teni :
    .\mathoff

when Appendix G, Rule 14 says:

    Otherwise if the font information shows a kern between the current
    symbol and the next, insert a kern item after the current Ord atom
    and move to the next item after that. Otherwise (i.e., if no
    ligature or kern is specified between the present text symbol and
    the following character), go to Rule 17.

and Rule 17 is the only place where italic corrections are added. The
                 [ ^^^^ also, \fontdimen 2\teni is zero, so we get
                   ital correction in spite of being marked as text -- DEK ]
phrasing `move to the next item after that' makes it sound as though
processing for the current item is complete. In fact, one should always
go to Rule 17 (this applies to ligatures as well, for which the phrasing
is now, since TeX 3.0 changes, quite ambiguous in Rule 14). The code
has always actually worked like that, and in fact it wouldn't make sense
to abandon the Ord before the kern at this stage, as its translation
into a horizontal list, needed for the second pass, has not yet been
generated.

Chris Thompson
-------
                    [ $2.56  right, I'll try to clean this up too -- DEK ]

		*****	 GFtype/ GF format	*****

Date: 18 September 1990, 16:34:53 GMT
From: PEB%DM0MPI11.BITNET@CUNYVM.CUNY.EDU
From: Peter Breitenlohner       (089) 32308-412      PEB      at DM0MPI11

Barbara, can you please relay the following to Don Knuth. Thanks P.B

Subject: Clarification of GF format and proposed GFtype change.

If I interpret the GF format correctly, it is an error if there
are raster data for a character c (mod 256) and there is no character
locator for character c in the postamble. It is also an error if
there are two or more locators for the same character.
(At least GFtoPK treats these cases as errors.)

If this is correct then it should be specified explicitely in the
definition of the GF format (in MF, GFtype, GFtoDVI, and GFtoPK).
                    [ I believe it is explicit enough, and
                      fixing GFtype will clinch it  -- DEK ]
Moreover GFtype should test for such anomalies. I find it extremely
embarrassing that GFtoPK declares such a file as bad whereas GFtype
[ ^^^^^^^^^^ Well, I am even more embarrassed, since I wrote GFtype  -- DEK ]
does not detect anything unusual.

Another point: GFtype cannot handle specials (xxx or yyy) in the
postamble. Clearly MF does not produce them but they are allowed in
the GF format. This is, however, slightly more tedious to fix.
               [ mistake in documentation.  They are allowed within
                 characters but not before _pre_ or after _post_  -- DEK ]

Therefore I propose the following change in GFtype.web (line numbers refe
to Version 3.0 as of November 5, 1989).

@x [8] m.64 l.1213 - test for missing character locators
@.should be postpost@>
@y
@.should be postpost@>
for k:=0 to 255 do if char_ptr[k]>0 then
  error('missing locator for character ',k:1,'!');
@.missing locator for character...@>
@z
%---------------------------------------
@x [8] m.64 l.1228 - superfluous semicolon
  error('not enough signature bytes at end of file!');
@y
  error('not enough signature bytes at end of file!')
@z
%---------------------------------------
@x [8] m.65 l.1250 - test for duplicate character locators
  if p<>char_ptr[c] then
    error('character location should be ',char_ptr[c]:1,'!');
@.character location should be...@>
@y
  if char_ptr[c]=0 then
    error('duplicate locator for this character!')
@.duplicate locator for this...@>
  else if p<>char_ptr[c] then
    error('character location should be ',char_ptr[c]:1,'!');
@.character location should be...@>
  char_ptr[c]:=0;
@z
%---------------------------------------
@x [8] m.65 l.1255 - superfluous semicolon
until k<>no_op;
@y
until k<>no_op
@z
%---------------------------------------
Regards   Peter Breitenlohner
-------

[ comment by DEK:
  Peter!
  I _really_ appreciate your help in getting these
  programs into tiptop shape.
  You are able to imitate my programming style
  better than I can myself -- don
]

		*****	MF.WEB, PK doc	*****

Date: Thu, 23 Aug 90 17:21:05 +0200
From: ef@tools.uucp (Edgar Fuss)
Subject: MF/PK Documentation errors

There is a documentation error (I think) in mf.web.
The section on scaled arithmetic says something like ``most internal
quantities MF deals with are supposed to be less than 4096 in absolute
value'' and then introduces the |fraction| type which has 28 binary digits
right to the binary point. In my opinion, on a 32-bit machine, you can only
represent values smaller than 4 in absolute value that way.
(However, the |angle| type introduced later on indeed has 20 digits of
precision and can hold values up to 2048-\epsilon.)
                                                     [ * ]

There is another bug in the description of the PK format that appears in
various WEB sources (PKtoPX, for example).
The definition of the |pre| command says that the |ds| parameter is the
                     [ ^^^ post  -- DEK ]
font's design size in units of 2^{-16} points. However, what GFtoPK really
puts there is the design size in units of 2^{-20} points.
This error really bit me last week (is my DVI the only device driver that
cares about design sizes and checksums?).
There are also a few errors in the same file where the meaning of various
flag bits are explained; the bit numbers and bit weights don't match,
I think, in genearal, the numbers are right and the weights are wrong.
                                                    [ ** ]

Edgar
-------

[ comment by DEK:
  * You are misinterpreting the comment in \S105.
    In \S101 it points out that _scaled_ numbers have 16 binary places
    If such numbers are < 2^{12}, that means we have 28 bits.
    To work with 28 significant bits its convenient to have a _fraction_
    type as well as a _scaled_ type.  Fractions have 28 binary places.
    Therefore `fractions' can get as large as $8-\epsilon$.
 ** I know nothing about PKtoPX (and I want PXL files abolished immediately)
    but I have been involved with GFtoPK, and it's documentation is
    currently correct ... sice November 89 at least. ...  Yes we had to
    fix it in many respects!  Please get up to date sources.
    Tom Rokicki's dvips driver is now extremely careful with
    design sizes and checksums, since I rewrote it in 1989.
]

		*****	PLtoTF/VPtoVF	*****

Date: Fri, 11 Jan 91 14:08:16 GMT
From: CET1@phoenix.cambridge.ac.uk
Subject: A PLtoTF/VPtoVF bug for Don from Wayne

Barbara,

Please could you forward the enclosed to Don Knuth?

> Accepted:  14:48:12 10 Jan 91
> From: "Wayne G. Sullivan" <WSULIVAN@EARN.IRLEARN>
> To: Chris Thompson <CET1@UK.AC.CAMBRIDGE.PHOENIX>
> Subject: A minor blemish in PLtoTF
>
> (DESIGNSIZE R 10.0)
> (DESIGNUNITS R 15.1)
> (FONTDIMEN
>    (SLANT R 0.00000000)
>    (SPACE R 10.0)
>    (XHEIGHT R 7.5)
>    (QUAD R 20.0)
>    (EXTRASPACE R 15.0))
> (CHARACTER C A
>    (CHARWD R 241.6)
>    )
>
> (COMMENT
> If the above file is fed to PLTOTF and the resulting TFM file is
> processed by TFTOPL the result is
> (FAMILY UNSPECIFIED)
> (FACE F MRR)
> (CODINGSCHEME UNSPECIFIED)
> (DESIGNSIZE R 10.0)
> (COMMENT DESIGNSIZE IS IN POINTS)
> (COMMENT OTHER SIZES ARE MULTIPLES OF DESIGNSIZE)
> (CHECKSUM O 32455153636)
> (SEVENBITSAFEFLAG TRUE)
> (FONTDIMEN
>    (SLANT R 0.0)
>    (SPACE R 0.662251)
>    (STRETCH R 0.0)
>    (SHRINK R 0.0)
>    (XHEIGHT R 0.496689)
>    (QUAD R 1.324503)
>    (EXTRASPACE R 0.993378)
>    )
> (CHARACTER C A
>    (CHARWD R 0.0)
>    )
> Note that the character width is zero, but no error message is
> given. This is machine dependent because of floating point
> arithemtic, but the computed value comes from 253335962 / 15833498
> which is less than 16.0 but rounds to the scaled equivalent of 16.0.
> )
 -----------------------------------------------------------------------
> Submitted: 15:28:45 10 Jan 91
> From: Chris Thompson <CET1>
> To: "Wayne G. Sullivan" <WSULIVAN@EARN.IRLEARN>
> Subject: Re: [A minor blemish in PLtoTF]
>
> Yes, agreed. |out_scaled| is defective in that the failure of test
> |abs(x/design_units)>=16.0| does not necessarily mean that
> |round((x/design_units)*1048576.0)| will deliver a value strictly
> less than 2^24. In your case there is a subsequent call of |out(256)|,
> which will blow up or quietly produce zero as compilation options or
> runtime routines determine.
>
> Shall I send it to Barbara for inclusion on Don's work list?
 -----------------------------------------------------------------------
> Submitted: 09:50:34 11 Jan 91
> From: "Wayne G. Sullivan" <WSULIVAN@EARN.IRLEARN>
> To: Chris Thompson <CET1@UK.AC.CAMBRIDGE.PHOENIX>
> Subject: out_scaled
>
> I believe the same out_scaled routine apears in VPtoVF. The problem with
> 16.0 is highly unlikely to cause problems in practice, but since these
> WEB files are considered as patterns of good software, it would seem to
> be in order to sort out the problem whenever there is a revised version of
> PLtoTF or VPtoVF. I would appreciate it if you would pass it on in the
> proper manner.
> I noticed the 16.0 problem when I was trying to figure out how 'shorten'
> works in connection with using PostScript resident fonts with TeX. The
> limitations of TFM files can have side effects. In the new character
> mapping for 256 character fonts given in Ferguson's article (TUG-11/4)
> a zero width character for inhibiting ligatures is included. It is
> possible, though unlikely, that after 'shorten' has been applied, that
> this character will no longer have zero width. It might be a good idea
> to make the 'rounding' message a bit more forceful to encourage further
> checks when rounding of character dimensions is necessary.
>                                                         Wayne

Chris Thompson
JANET:    cet1@uk.ac.cam.phx
Internet: cet1%phx.cam.ac.uk@nsfnet-relay.ac.uk
-------

[ comment by DEK:
  I guess you're right, but only if all 256 widths are different
  _and_ the closest two widths happen to be between 0 and something
  very near 0!
  The main side effects I've ever noticed were in connection with
  math extension font ... it is very important not to diddly with
  height or depth of characters used to build up large parens etc.

    PLtoTF version 3.4   \  have out_scaled
    VPtoVF version 1.3   /  more robust!
                    Thanks again  don
            ?
  Interestingly if you had had R-241.6
  the output would have contained R-16.0
]

			*****	TANGLE	*****

Date: Fri, 01 Mar 91 13:08:38 GMT
From: Peter Breitenlohner <PEB%DM0MPI11.BITNET@CUNYVM.CUNY.EDU>
Organization: Max-Planck-Institut fuer Physik, Muenchen
Subject: patgen and tangle

[ while i was in vienna at the dante meeting, i talked with peter about
  the maintenance of various ur-tex-utilities, in particular patgen.
  i couldn't say who, if anyone, is currently responsible for patgen
  maintenance.  can you shed any light on this? ]
          [ This has been up for grabs ever since Frank finished his
            thesis.  All volunteers welcome to step up and advertise
            their presence.  -- DEK ]

I found a bug in tangle. If there is a fatal error in phase on (i.e.,
while digesting the input), the token memory usage statistic is garbage.
This is particularly annoying if the fatal error is token memory overflow
(as it happened to me).

Below is my proposed bug fix, line numbers refer to
      TANGLE.WEB 4.1 as of September 5, 1990.

@x [18] m.186 l.3272 - bug fix: initialize max_tok_ptr
print(' bytes, ', max_tok_ptr[0]:1);
@y
if phase_one then
  for ii:=0 to zz-1 do max_tok_ptr[ii]:=tok_ptr[ii];
print(' bytes, ', max_tok_ptr[0]:1);
@z
%---------------------------------------

Regards Peter
-------

[ comment by DEK:
  Thanks Peter for excellent diagnosis & correction.
  Will be fixed in v4.2.  -- don
]


########################################################################

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%  Character code reference
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%                       Upper case letters: ABCDEFGHIJKLMNOPQRSTUVWXYZ
%                       Lower case letters: abcdefghijklmnopqrstuvwxyz
%                                   Digits: 0123456789
% Square, curly, angle braces, parentheses: [] {} <> ()
%           Backslash, slash, vertical bar: \ / |
%                              Punctuation: . ? ! , : ;
%          Underscore, hyphen, equals sign: _ - =
%                Quotes--right left double: ' ` "
%"at", "number" "dollar", "percent", "and": @ # $ % &
%           "hat", "star", "plus", "tilde": ^ * + ~
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

[ end of message 030 ]
-------