1
% The TRAP manual: How to validate MF --- last updated by D E Knuth on 4 Dec 89
4
\font\titlefont=cmssdc10 at 40pt
6
\font\logo=manfnt % font used for the METAFONT logo
7
\def\MF{{\logo META}\-{\logo FONT}}
11
\def\.#1{\hbox{\tt#1}}
12
\def\\#1{\hbox{\it#1\/\hskip.05em}} % italic type for identifiers
15
\baselineskip 12pt plus .25pt
17
\def\verbatim#1{\begingroup \frenchspacing
18
\def\do##1{\catcode`##1=12 } \dospecials
19
\parskip 0pt \parindent 0pt
20
\catcode`\ =\active \catcode`\^^M=\active
21
\tt \def\par{\ \endgraf} \obeylines \obeyspaces
23
% a blank line will be typeset at the end of the file;
24
% if you're unlucky it will appear on a page by itself!
25
{\obeyspaces\global\let =\ }
27
\output{\shipout\box255\global\advance\pageno by 1} % for the title page only
30
\centerline{\titlefont A Torture Test}
32
\centerline{\titlefont for \logo ()*+,-.*}
34
\centerline{by Donald E. Knuth}
35
\centerline{Stanford University}
37
\centerline{({\sl Version 2, January 1990\/})}
39
\centerline{\vbox{\hsize 4in
40
\noindent Programs that claim to be implementations of \MF84 are
41
supposed to be able to process the test routine contained in this
42
report, producing the outputs contained in this report.}}
46
The preparation of this report was supported in part by the National Science
47
Foundation under grants IST-8201926 and MCS-8300984,
48
and by the System Development Foundation.
49
{\logo opqrstuq} is a trademark of Addison-Wesley Publishing Company.
54
\output{\shipout\vbox{ % for subsequent pages
55
\baselineskip0pt\lineskip0pt
57
\ifodd\pageno \hfil\eightrm\firstmark\hfil
59
\else\mainfont\the\pageno\hfil
60
\eightrm\firstmark\hfil\fi}
63
\global\advance\pageno by 1}
64
\let\runninghead=\mark
65
\outer\def\section#1.{\noindent{\bf#1.}\quad
66
\runninghead{\uppercase{#1} }\ignorespaces}
68
\section Introduction.
69
People often think that their programs are ``debugged'' when large applications
70
have been run successfully. But system programmers know that a typical large
71
application tends to use at most about 50 per cent of the instructions
72
in a typical compiler. Although the other half of the code---which tends
73
to be the ``harder half''---might be riddled with errors, the system seems
74
to be working quite impressively until an unusual case shows up on the
75
next day. And on the following day another error manifests itself, and so on;
76
months or years go by before certain parts of the compiler are even
77
activated, much less tested in combination with other portions of the system,
78
if user applications provide the only tests.
80
How then shall we go about testing a compiler? Ideally we would like to
81
have a formal proof of correctness, certified by a computer.
82
This would give us a lot of confidence,
83
although of course the formal verification program might itself be incorrect.
84
A more serious drawback of automatic verification is that the formal
85
specifications of the compiler are likely to be wrong, since they aren't
86
much easier to write than the compiler itself. Alternatively, we can
87
substitute an informal proof of correctness: The programmer writes his or
88
her code in a structured manner and checks that appropriate relations
89
remain invariant, etc. This helps greatly to reduce errors, but it cannot
90
be expected to remove them completely; the task of checking a large
91
system is sufficiently formidable that human beings cannot do it without
92
making at least a few slips here and there.
94
Thus, we have seen that test programs are unsatisfactory if they are simply
95
large user applications; yet some sort of test program is needed because
96
proofs of correctness aren't adequate either. People have proposed schemes
97
for constructing test data automatically from a program text, but such
98
approaches run the risk of circularity, since they cannot assume that a
99
given program has the right structure.
101
I have been having good luck with a somewhat different approach,
102
first used in 1960 to debug an {\mc ALGOL} compiler. The idea is to
103
construct a test file that is about as different from a typical user
104
application as could be imagined. Instead of testing things that people
105
normally want to do, the file tests complicated things that people would
106
never dare to think of, and it embeds these complexities in still
107
more arcane constructions. Instead of trying to make the compiler do the
108
right thing, the goal is to make it fail (until the bugs have all been found).
110
To write such a fiendish test routine, one simply gets into a nasty frame
111
of mind and tries to do everything in the unexpected way. Parameters
112
that are normally positive are set negative or zero; borderline cases
113
are pushed to the limit; deliberate errors are made in hopes that the
114
compiler will not be able to recover properly from them.
116
A user's application tends to exercise 50\%\ of a compiler's logic,
117
but my first fiendish tests tend to improve this to about 90\%. As the
118
next step I generally make use of frequency-counting software to identify
119
the instructions that have still not been called upon. Then I add ever more
120
fiendishness to the test routine, until more than 99\%\ of the code
121
has been used at least once. (The remaining bits are things that
122
can occur only if the source program is really huge, or if certain
123
fatal errors are detected; or they are cases so similar to other well-tested
124
things that there can be little doubt of their validity.)
126
Of course, this is not guaranteed to work. But my experience in 1960 was
127
that only two bugs were ever found in that {\mc ALGOL} compiler after it
128
correctly translated that original fiendish test. And one of those bugs
129
was actually present in the results of the test; I simply had failed to
130
notice that the output was incorrect. Similar experiences occurred later
131
during the 60s and 70s, with respect to a few assemblers, compilers,
132
and simulators that I wrote.
134
This method of debugging, combined with the methodology of structured
135
programming and informal proofs (otherwise known as careful desk checking),
136
leads to greater reliability of production software than any other
137
method I know. Therefore I have used it in developing \MF84, and the
138
main bulk of this report is simply a presentation of the test program
139
that was used to get the bugs out of \MF.
141
Such a test file is useful also after a program has been debugged, since
142
it can be used to give some assurance that subsequent modifications don't
145
The test file is called \.{TRAP.MF}, because of my warped sense of humor:
146
\MF's companion system, \TeX, has a similar test file called \.{TRIP}, and I
147
couldn't help thinking about Billy Goat Gruff and the story of ``trip,
150
The contents of this test file are so remote from what people actually
151
do with \MF, I feel apologetic if I have to explain the correct
152
translation of \.{TRAP.MF}; nobody really cares about most of the
153
nitty-gritty rules that are involved. Yet I believe \.{TRAP} exemplifies
154
the sort of test program that has outstanding diagnostic ability, as
157
If somebody claims to have a correct implementation of \MF, I will not
158
believe it until I see that \.{TRAP.MF} is translated properly.
159
I propose, in fact, that a program must meet two criteria before it
160
can justifiably be called \MF: (1)~The person who wrote it must be
161
happy with the way it works at his or her installation; and (2)~the
162
program must produce the correct results from \.{TRAP.MF}.
164
\MF\ is in the public domain, and its algorithms are published;
165
I've done this since I do not want to discourage its use by placing
166
proprietary restrictions on the software. However, I don't want
167
faulty imitations to masquerade as \MF\ processors, since users
168
want \MF\ to produce identical results on different machines.
169
Hence I am planning to do whatever I can to suppress any systems that
170
call themselves \MF\ without meeting conditions (1) and~(2).
171
I have copyrighted the programs so that I have some chance to forbid
172
unauthorized copies; I explicitly authorize copying of correct
173
\MF\ implementations, and not of incorrect ones!
175
The remainder of this report consists of appendices, whose contents ought
176
to be described briefly here:
178
Appendix A explains in detail how to carry out a test of \MF, given
179
a tape that contains copies of the other appendices.
181
Appendix B is \.{TRAP.MF}, the fiendish test file that has already
182
been mentioned. People who think that they understand \MF\ are challenged
183
to see if they know what \MF\ is supposed to do with this file.
184
People who know only a little about \MF\ might still find it
185
interesting to study Appendix~B, just to get some insights into the
186
methodology advocated here.
188
Appendix C is \.{TRAPIN.LOG}, a correct transcript file \.{TRAP.LOG}
189
that results if \.{INIMF} is applied to \.{TRAP.MF}. (\.{INIMF} is
190
the name of a version of \MF\ that does certain initializations;
191
this run of \.{INIMF} also creates a binary base file called \.{TRAP.BASE}.)
193
Appendix D is a correct transcript file \.{TRAP.LOG} that results if
194
\.{INIMF} or any other version of \MF\ is applied to \.{TRAP.MF}
195
with base file \.{TRAP.BASE}.
197
Appendix E is \.{TRAP.TYP}, the symbolic version of a correct output
198
file \.{TRAP.72270GF} that was produced at the same time as the \.{TRAP.LOG}
201
Appendix F is \.{TRAP.PL}, the symbolic version of a correct output
202
file \.{TRAP.TFM} that was produced at the same time as the \.{TRAP.LOG}
205
Appendix G is \.{TRAP.FOT}, an abbreviated version of Appendix D that
206
appears on the user's terminal during the run that produces \.{TRAP.LOG},
207
\.{TRAP.72270GF}, and \.{TRAP.TFM}.
209
The debugging of \MF\ and the testing of the adequacy of \.{TRAP.MF}
210
could not have been done nearly as well as reported here except for
211
the magnificent software support provided by my colleague David R. Fuchs.
212
In particular, he extended our local Pascal compiler so that
213
frequency counting and a number of other important features were added
214
to its online debugging abilities.
216
The method of testing advocated here has one chief difficulty that deserves
217
comment: I had to verify by hand that \MF\ did the right things
218
to \.{TRAP.MF}. This took many hours, and perhaps I have missed
219
something (as I did in 1960); I must confess that I have not checked
220
every single number in Appendices D, E, and~F. However, I'm willing to pay
221
$\$$81.92 to the first finder of any remaining bug in \MF, and I will
222
be surprised if that bug doesn't show up also in one of these appendices.
226
\section Appendix A: How to test \MF.
228
\item{0.} Let's assume that you have a tape containing \.{TRAP.MF},
229
\.{TRAPIN.LOG}, \.{TRAP.LOG}, \.{TRAP.TYP}, \.{TRAP.PL}, and \.{TRAP.FOT},
230
as in Appendices B, C, D, E, F, and~G. Furthermore, let's suppose that you
231
have a working \.{WEB} system, and that you have working programs
232
\.{TFtoPL} and \.{GFtype}, as described in the \TeX ware and \MF ware reports.
234
\item{1.} Prepare a version of \.{INIMF}. (This means that your \.{WEB}
235
change file should have {\bf init} and {\bf tini} defined to be null.)
236
The {\bf debug} and {\bf gubed} macros should be null, in order to
237
activate special printouts that occur when $\\{tracingedges}>1.0$.
238
The {\bf stat} and {\bf tats} macros should also be null, so that
239
statistics are kept. Set \\{mem\_top} and \\{mem\_max} to 3000
240
(or to \\{mem\_min} plus 3000, if \\{mem\_min} isn't zero),
241
for purposes of this test version.
242
Also set $\\{error\_line}=64$, $\\{half\_error\_line}=32$,
243
$\\{max\_print\_line}=72$, $\\{screen\_width}=100$, and
244
$\\{screen\_depth}=200$; these parameters affect many of the lines of
245
the test output, so your job will be much easier if you use the same
246
settings that were used to produce Appendix~E. Also (if possible) set
247
$\\{gf\_buf\_size}=8$, since this tests more parts of the program.
248
You probably should also use the ``normal'' settings of other parameters
249
found in \.{MF.WEB} (e.g., $\\{max\_internal}=100$, $\\{buf\_size}=500$,
250
etc.), since these show up in a few lines of the test output. Finally,
251
change \MF's screen-display routines by putting the following simple lines
253
$$\vbox{\halign{\tt#\hfil\cr
254
\char`\@x Screen routines:\cr
255
begin init\char`\_screen:=false;\cr
257
begin init\char`\_screen:=true;
258
\char`\{screen instructions will be logged\char`\}\cr
260
None of the other screen routines (\\{update\_screen}, \\{blank\_rectangle},
261
\\{paint\_row}) should be changed in any way; the effect will be to have
262
\MF's actions recorded in the transcript files instead of on the screen,
263
in a machine-independent way.
265
\item{2.} Run the \.{INIMF} prepared in step 1. In response to the first
266
`\.{**}' prompt, type carriage return (thus getting another `\.{**}').
267
Then type `\.{\char`\\input trap}'. You should get an output that matches
268
the file \.{TRAPIN.LOG} (Appendix~C). Don't be alarmed by the error
269
messages that you see, unless they are different from those in Appendix~C.
272
\item{3.} Run \.{INIMF} again. This time type `\.{\sp\&trap\sp\sp trap\sp}'.
273
(The spaces in this input help to check certain parts of \MF\ that
274
aren't otherwise used.) You should get outputs \.{TRAP.LOG}, \.{TRAP.72270GF},
276
Furthermore, your terminal should receive output that matches \.{TRAP.FOT}
277
(Appendix~G). During the middle part of this test, however, the terminal
278
will not be getting output, because \.{batchmode} is being
279
tested; don't worry if nothing seems to be happening for a while---nothing
282
\item{4.} Compare the \.{TRAP.LOG} file from step 3 with the ``master''
283
\.{TRAP.LOG} file of step~0. (Let's hope you put that master file in a
284
safe place so that it wouldn't be clobbered.) There should be perfect
285
agreement between these files except in the following respects:
287
\itemitem{a)} The dates and possibly the file names will
288
naturally be different.
290
\itemitem{b)} If you had different values for \\{stack\_size}, \\{buf\_size},
291
etc., the corresponding capacity values will be different when they
292
are printed out at the end.
294
\itemitem{c)} Help messages may be different; indeed, the author encourages
295
non-English help messages in versions of \MF\ for people who don't
296
understand English as well as some other language.
298
\itemitem{d)} The total number and length of strings at the end and/or
299
``still untouched'' may well be different.
301
\itemitem{e)} If your \MF\ uses a different memory allocation or
302
packing scheme, the memory usage statistics may change.
304
\itemitem{f)} If you use a different storage allocation scheme, the
305
capsule numbers will probably be different, but the order of variables
306
should be unchanged when dependent variables are shown. \MF\ should also
307
choose the same variables to be dependent.
309
\itemitem{g)} If your computer handles integer division of negative operands
310
in a nonstandard way, you may get results that are rounded differently.
311
Although \TeX\ is careful to be machine-independent in this regard,
312
\MF\ is not, because integer divisions are present in so many places.
314
\item{5.} Use \.{GFtype} to convert your file \.{TRAP.72270GF} to a file
315
\.{TRAP.TYP}. (Both of \.{GFtype}'s options, i.e., mnemonic output and image
316
output, should be enabled so that you get the maximum amount of output.)
317
The resulting file should agree with the master \.{TRAP.TYP} file of step~0,
318
assuming that your \.{GFtype} has the ``normal'' values of compile-time
319
constants ($\\{top\_pixel}=69$, etc.).
321
\item{6.} Use \.{TFtoPL} to convert your file \.{TRAP.TFM} to a file
322
\.{TRAP.PL}. The resulting file should agree with the master \.{TRAP.PL}
325
\item{7.} You might also wish to test \.{TRAP} with other versions of
326
\MF\ (i.e., \.{VIRMF} or a production version with another base file
327
preloaded). It should work unless \MF's primitives have been redefined in
328
the base file. However, this step isn't essential, since all the code of
329
\.{VIRMF} appears in \.{INIMF}; you probably won't catch any more errors
330
this way, unless they would already become obvious from normal use of
335
\section Appendix B: The \.{TRAP.MF} file.
336
The contents of the test routine are prefixed here with line numbers, for
337
ease in comparing this file with the error messages printed later; the
338
line numbers aren't actually present.
339
\runninghead{APPENDIX B: \.{TRAP.MF} (CONTINUED)}
342
\begingroup\count255=0
343
\everypar{\global\advance\count255 by 1
344
\hbox to 20pt{\sevenrm\hfil\the\count255\ \ }}
349
\section Appendix C: The \.{TRAPIN.LOG} file.
350
When \.{INIMF} makes the \.{TRAP.BASE} file, it also creates a file called
351
\.{TRAP.LOG} that looks like this.
352
\runninghead{APPENDIX C: \.{TRAPIN.LOG} (CONTINUED)}
355
\verbatim{trapin.log}
358
\section Appendix D: The \.{TRAP.LOG} file.
359
Here is the major output of the \.{TRAP} test; it is generated by running
360
\.{INIMF} and loading \.{TRAP.BASE}, then reading \.{TRAP.MF}.
361
\runninghead{APPENDIX D: \.{TRAP.LOG} (CONTINUED)}
363
{\let\tt=\eighttt\leftskip 1in\baselineskip 9pt plus .1pt minus .1pt
369
\section Appendix E: The \.{TRAP.TYP} file.
370
Here is another major component of the test. It shows the output of \.{GFtype}
371
applied to the file \.{TRAP.72270GF} that is created at the same time
372
Appendix D was produced.
373
\runninghead{APPENDIX E: \.{TRAP.TYP} (CONTINUED)}
375
{\let\tt=\eighttt\leftskip 1in\baselineskip 9pt plus .1pt minus .1pt
381
\section Appendix F: The \.{TRAP.PL} file.
382
In this case we have the output of \.{TFtoPL}
383
applied to the file \.{TRAP.TFM} that is created at the same time
384
Appendix D was produced.
385
\runninghead{APPENDIX F: \.{TRAP.PL} (CONTINUED)}
387
{\let\tt=\eighttt\leftskip 1in\baselineskip 9pt plus .1pt minus .1pt
393
\section Appendix G: The \.{TRAP.FOT} file.
394
This shows what appeared on the terminal while Appendix D was being produced.
395
\runninghead{APPENDIX G: \.{TRAP.FOT} (CONTINUED)}