§11. Excerpts from class, October 21
[notes by PMR]
Don opened class with the good news that Mary-Claire van Leunen has agreed to helpread the term papers and drafts thereof, despite the fact that her name was incorrectlycapitalized in last week’s notes.
Returning to the subject of “Literate Programming,” Don said that it takes a while
to ﬁnd a new style to suit a new system like WEB. When he was trying to write the WEBprogram in its own language he tore up his ﬁrst 25 pages of code and started again, havingﬁnally found a comfortable style. He digressed to talk about the vicious circle involved inwriting a program in its own language. To break it, he hand-simulated the program onitself to produce a Pascal program that could then be used to compile WEB programs. Thetask was eased because there is obviously no need for error-handling routines when dealingwith code that you have to debug anyway. But there is also another kind of bootstrappinggoing on; you can evolve a style to write these programs only by sitting down and writingprograms. Don told us that he wrote WEB in just two months, as it was never intended tobe a polished product like TEX.
We spent the rest of the class looking at WEB programs that had been written by
undergraduates doing independent research with Don during the Spring. We saw howthey had (or had not) adapted to its style. Don said that he had got a lot of feedbackand sometimes found it hard to be dispassionate about stylistic questions, but that somethings were clearly wrong. He showed us an example that looked for all the world just likea Pascal program; the student had obviously not changed his ways of thinking or writingat all, and so had failed to make any use of the features of the system. The English in hisintroductory paragraph also left a lot to be desired.
Don showed us his thick book TEX: The Program
—a listing of the code for TEX,
written in WEB. It consists of almost 1400 modules. The guiding principle behind WEB isthat each module is introduced at the psychologically right moment. This means that theprogram can be written in such a way as to motivate the reader, leaving TANGLE to sorteverything out later on. [The TANGLE processor converts WEB programs to Pascal programs.]After all, we don’t need to worry about motivating the compiler. (Don added the asidethat contrary to superstition, the machine doesn’t spend most of its time executing thoseparts of the code that took us the longest to write.) It seems to be true that the best wayin which to present program constructs to the reader is to use the same order in which thecreator of the program found himself making decisions about them. Don himself alwaysfelt it was quite clear what had to be presented next, throughout the entire compositionof this huge program. There was at all points a natural order of exposition, and it seemsthat the natural orderings for reading and writing are very much the same.
The ﬁrst student hadn’t used this new ﬂexibility at all; he had essentially just used
WEB to throw in comments here and there.
A general problem of exposition arose: How are we to describe the behavior of a
computer program? Do we see the program as essentially autonomous, “running itself,” orare we participants in the action? Our attitude to this determines whether we are goingto say ‘we insert the element in the heap’ or ‘it inserts the element . . .
’. Don favours ‘we’;at any rate one should be consistent.
Students used descriptors and imperatives for the names of their modules; Don said
he favours the latter, as in 〈
Store the word in the dictionary 〉
, which works much betterthan 〈
Stores the word in the dictionary 〉
. On the other hand, where a module is essentiallya piece of text with a declarative function—a list of declarations, say—we should use adescriptor to name it: 〈
Procedures for sorting 〉
Incidentally, it is natural to capitalize the ﬁrst letter of a module name.
One student used the identiﬁer ‘FindInNewWords
’. This looks comparatively bad
in print: Uppercase letters were not designed to appear immediately following lowercaseones. Since the use of compound nouns is almost inevitable, WEB provides a neat solution.
It allows a short underscore to be used to conjoin words like get word
. (Since the Pascalcompiler will not accept identiﬁers like this, TANGLE quietly removes the underscore.) Dontold us that Jim Dunlap of Digitek, who made some of the best early compilers, invariablyused identiﬁers forty-or-so characters long. The meaning was always quite clear althoughno comments appeared.
Each module should contain an informal but clear description of what it actually does.
A play-by-play account of an algorithm, a simple stepping through of the process, does notqualify. We are trying to convey an intuition of what is going on, so a high-level accountis much more helpful.
We saw several modules that were much too long. Don thinks that a dozen lines
of code is about the right length for a module. Often he simply recommended that thestudents cut the oﬀending specimens into several pieces, each of more manageable size.
The whole philosophy behind WEB is to break a complex thing into tractable parts, so thecode should reﬂect this. Once you get the idea, you begin writing code this way, and it’seasier.
We saw an example in which the student had slipped into “engineerese” in his de-
scriptive text—all conjunctions and no punctuation. This worked for James Joyce, but
it doesn’t make for good documentation. One student had apparently managed to break
WEB—the formatting of begin
s and end
s came out all wrong. Heaven knows what he did.
One student put comments after each end
to show what was being ended, as end
. This is a good idea when writing Pascal, but it’s unnecessary in WEB. Thus it’s
a good example of a convention that is no longer appropriate to the new style; when you
change style you needn’t carry excess baggage along.
Don had more to say about the anthropomorphization of computer systems. Why
prompt the user with ‘Name of file to process?’ when we can have the computer say‘What file should I process?’ ? Don generally likes the use of ‘I’ by the computerwhen referring to itself, and thinks this makes it easier for users to conceptualize whatis going on. Perhaps humans can think of complex processes best in terms of demons inboxes, so why not acknowledge this? Eliza, the AI program that simulates a certain typeof psychiatrist, managed to fool virtually everyone by an extension of this approach. Elizamay or may not be a recommendation for anthropomorphisms, or for psychiatry. Thereare those, such as Dijkstra, who think such use of ‘I’ to be a bad thing.
As in the case of maths, don’t start a sentence with a symbol. So don’t say ‘data
assumes that . . .
’—it can easily be rewritten.
We saw several programs by one student who had developed a very distinctive and
(Don thought) colourful style. His prose is littered with phrases like “Oooops! How can weﬁx this?” and “Now to get down to the nitty-gritty.” This stream-of-consciousness stylereally does seem to motivate reading, and helps infect the reader with the author’s obviousenthusiasm. There were a few small nits to pick with this guy though: His descriptionscould often be more descriptive. Why not call a variable caps range
instead of just range
?Don also had to point out to him that ‘complement’ and ‘component’ are in fact twodiﬀerent words.
In WEB you can declare your variables at any point in the program. Don thinks it is
always a good idea to add some comment when you do so, even if only a very cursoryexplanation is needed.
A note about asterisks: Be warned that typeset asterisks tend to appear higher above
the line than typewritten ones, so your multiplication formulæ may come out lookingstrange. Better to use ×
for multiplication, and to use a typewriter-style font with body-centered ‘*’ symbols instead of the ‘*’ in normal typographic fonts.
Another freshman was digitizing the Mona Lisa for reasons best known to insiders
Don pointed out that since the program uses a somewhat
specialized data structure (the heap) that might be unknown to the readers, the authorshould keep all the heap routines together in the text so that they can be read as a groupwhile fresh in the reader’s mind. In WEB we are not constrained by top-down, bottom-up,or any other order.
This student capitalized the ﬁrst letter of every word in titles of modules, even ‘And’
and the like. This looks rather unnatural—it is better to follow the newspaper-headlinesconvention by leaving such words entirely in lowercase, and even better to capitalize onlythe ﬁrst word.
Don thought it a good idea to use typewriter type for hexadecimal numbers, for
instance when saying ‘3F represents 63’. But leave the ‘63’ in normal type. This conventionlooks appropriate and provides a kind of subliminal type-checking.
The words used in the documentation should match the words used in the formal
program—you will only confuse the reader by using two diﬀerent terms for the same thing.
It’s a good idea to develop the habit of putting your begin
s and end
s inside the called
modules, not putting them in the calling module. That is, do it like this:
= 4 then 〈
= 4 then
Incidentally, appalling bugs will occur if we mix the two conventions!
DATOS PERSONALES Fecha Nacimiento: 13-12-59 (49 años) Lugar: Cdad Autónoma de Bs.As. Pcia. Bs.As. Argentina. Domicilio Particular: José Marti 368 – Ciudad Autónoma de Bs.As. Provincia de Bs.As . TE 4637-5971 TC 15-4411-4680 E-mail [email protected] – www.ginecologiaestetica.com.ar Consultorios: F. Lacroze 2563 1°A – CABA. TE 4555-1212/1313 Otamendi
Dourado VZ, Tanni SE, Vale SA, Faganello MM, Sanchez FF, Godoy I Artigo de Revisão Manifestações sistêmicas na doença pulmonar obstrutiva crônica* Systemic manifestations in chronic obstructive pulmonary disease VICTOR ZUNIGA DOURADO1, SUZANA ERICO TANNI2, SIMONE ALVES VALE2, MÁRCIA MARIA FAGANELLO1, FERNANDA FIGUEIRÔA SANCHEZ3, IRMA GODOY4 RESUMO A doença pulmonar obstrutiv