Quelles erreurs de débutant dois-je éviter?

LaTeX has a reputation for producing excellent results, but at the cost of a steep learning curve. That’s true, but by understanding a few basic principles, and learning how to avoid some techniques that may seem obvious but often lead one into the weeds, it’s possible to avoid some of that pain. Our goal here is to encourage good habits before bad habits have had a chance to develop.

The examples presented here are drawn from two main sources:

• In the author’s years as part of the TeXnical support team for a major math publisher, responsibilities included fielding questions from authors and writing user documentation.
• The online TeX forum at StackExchange1 has provided a surfeit of questions both basic and advanced. A community effort has collected a list of “Often referenced questions”, by topic, at tex.meta.stackexchange.com/q/2419.

There are several concepts that seem to be either missing from a new user’s bag of tricks, or not clearly understood. Let’s get them out of the way up front.

Many new (La)TeX users think that the document class is the template for a particular style or publication. Not so, although the thought is going in the right direction. A template is a source (.tex) file that is an outline. It begins with \documentclass and contains a minimum of basic structural commands into which additional definitions and text can be inserted as appropriate. Ideally, the template itself can be compiled with no errors resulting, but without producing any useful output.

Most new users these days enter (La)TeX from an editor or other GUI, and launch a non-interactive job that will blithely keep on process- ing the file until it finishes (with or without errors) or hangs in a loop. Launching the compilation from the command line, on the other hand, allows one to interact with the session and, in certain cases, make corrections “on the fly”, or if that’s not possible, halt the job in case of an error before the collection of reported errors becomes unhelpful. One type of “fixable” error is a misspelled command:

! Undefined control sequence.
l.37 \scetion
{Section}
?

Respond to this with the correct spelling;

i\section

hit “return”, and continue; don’t forget to correct the file when you come to a good stopping point. A misspelled environment name can’t be cor- rected this way; if that happens, cancel the job with an x, fix the file, and start over. Continuing a run after an unfixable error will just result in more er- ror messages, most of which are meaningless and confusing, so it’s best to avoid them.

Every time a TEX job runs, it will create a log file. Learn where to find this file! In addition to errors and warnings, it will report all files that were read in, including version numbers for document class files and packages, pages processed, and, at the end, resources used. Only a few relevant items will be mentioned here, but in a paper based on an earlier talk, instructions are given for how to undertake serious debugging.

In order to avoid overfull lines, error and warning messages shown here may be broken to fit the narrow columns of the TUGboat style. Many error messages output by LaTeX will consist of several lines, the first being the message, and the next showing the number of the line on which the error is identified along with the content of that line, up through the error text. A following line, indented so that it, with the numbered line, completes the line as it appears in the input.

Although we will deal here mostly with details, please remember that the basic concept of LaTeX is to separate content from structure.

Here we deal with some fundamentals of LaTeX.

Instructions are given to (La)TeX by means of commands, or “control sequences”, which by default begin with a backslash (\). There are two varieties: those which consist of the backslash followed by one non-letter character (“control symbol”), and those of one or more letters (“control words”) in which only letters (upper- or lowercase A–Z) are permitted (no digits or special characters). A control word may have one or more arguments (\title{…}) or stand by itself (\alpha). A “standalone” control word will be terminated by a space or any other non-letter. But a space after a control symbol will appear as a space in the output. Several control symbols are predefined to produce their own character in the output: \#, \%, \$, \&. For example, \$ produces .

A user can define new commands, or assign new meanings to existing commands. LaTeX provides \newcommand to create a brand-new definition. \newcommand checks to make sure that the command name hasn’t been used before, and complains if it has. (The basic TeX \def does not.) If it’s necessary to redefine a command that already exists, the recommended way is to use \renewcommand — but be sure you know what you’re doing. For example, redefining \par is chancy, as LaTeX uses this “under the covers” for many different formatting adjustments, and it’s very easy to mess things up.

Single-letter commands are also bad candidates for (re)definition by users, as many of them are predefined as accents or forms of letters not usual in English text; \i might very well occur with (or without) an accent in a references list. For (a bad) example, consider the author Haïm Brezis:

\renewcommand{\i}{\ensuremath{\sqrt{-1}}}
Brezis, Ha\"{\i}m

Single-digit commands (\0, \1, etc.) are not predefined in core LaTeX, so are available for ad hoc use.

An environment is a block of material between

\begin{⟨env-name⟩}
...
\end{⟨env-name⟩}.

The environment name must match at beginning and end; if it doesn’t, this error is reported in the log file and on the terminal:

! LaTeX Error: \begin{xxx} on input line nn
ended by \end{yyy}

Most environments can be nested, but the proper sequence must be maintained.

Other commands are available to provide new definitions — \NewDocumentCommand, \NewEnvironment, \NewDocumentEnvironment and similar ones for redefinitions. For details on these, consult a current reference.

Generally speaking, the current mode identifies where you are on the (output) page, but here we will take a point of view based on the input/source file.

There are three modes: vertical, horizontal and math.

Starting after \documentclass or after a blank line or an explicit \par, LaTeX is in vertical mode. Certain operations are best launched in vertical mode; more about this later.

Starting to input ordinary text is one way to enter horizontal mode. Other transitions from vertical to horizontal mode are \indent, \noindent and \leavevmode. Within horizontal mode, multiple consecutive spaces are treated as a single space; consecutive is essential here. An end-of-line (EOL) is treated as a space, even though it’s not explicitly visible in the source file; a GUI that wraps lines may or may not (usually not) insert an EOL, and different operating systems define an EOL differently, but such differences are taken care of by the TEX engine. Spaces at the beginning of a line are ignored. More about spaces later on.

The third mode, math, can be embedded in-line in text or set as display material in vertical mode. Inline math is wrapped in $ signs or surrounded by \(…\). An unnumbered one-line display can be indicated by \[…\]. Multi-line math displays are best entered using the environments provided by the amsmath and mathtools packages. (Refer to the user documentation; mathtools loads amsmath, so it’s not necessary to load both.) A math display is usually a continuation of the preceding paragraph, so don’t leave a blank line between a display and the preceding text; to do so can result in an unwanted page break.

Within math mode, blank lines are not allowed; this was a decision made by Knuth, to catch unintentional input lapses, since math never continues across a paragraph break.

Along with modes, there is the concept of scope, making it possible to localize definitions and operations.

Math mode is one instance of scope; certain characters and operations are valid only within math, and others are invalid there. Within text, math usually begins and ends with $, and these must be matched. Display math breaks the flow of text; closing a display returns to text mode unless followed by a blank line or \par. More about math later. Another way of delimiting scope is to wrap the material in braces: {…}. Within this scope, the meaning of a command may be changed for temporary effect; the definition in effect before the opening brace will be restored as soon as the closing brace is digested. Instead of a brace pair, the commands \begingroup…\endgroup have the same effect.

Another way to have a scoped environment is to pack the material in a “box”. This may be a minipage, \mbox or \parbox. Other boxes are defined in packages like tcolorbox. Some environments (not all) are defined to be a scope. One such is the theorem environment, inside which text is italic; when the theorem ends, the text style automatically reverts to the document default.

A goal of high-quality typesetting is even spacing in text. This is really possible only with ragged-right setting, where spaces are “natural width”. But even margins are usually preferred, so TeX is designed to optimize spacing in that context.

In U.S. documents, spaces that end sentences are wider than interword spaces. This is not true for documents in other languages, and can be turned off with \frenchspacing. But in academic documents, frequent abbreviations can make it difficult to tell where sentences end. To avoid a too-wide space after an abbreviation, follow it by \␣ (backslash-space):

abc vs. xyz
 
abc vs.\ xyz

If the line shouldn’t break after the abbreviation, follow the period by ~:

seen on p.~23.

A similar, but reverse, situation can occur when an uppercase letter is followed by a period. This is assumed to be the initial of a name; it usually is, and an ordinary interword space is set. But sometimes the uppercase letter is at the end of an acronym, and that ends a sentence. In such a case, add \@ before the period, and it will restore the wider end-of-sentence space.

All this boils down to a simple rule: Except at the end of a sentence (and to a lesser extent after other punctuation symbols or within math), all spaces within the same line should be the same width. If they’re not, something is fishy.

Remember that a space terminates a control word and it’s then discarded; that’s one place where it’s not necessary to input a %. But there are places where adding a % can really cause trouble.

After defining any numeric value, TeX will keep looking for anything else that can be interpreted as numeric, so if a line ends with \xyz=123, no % should be added. Or, if setting a rubber length (glue), say \parskip=2pc, TeX will keep looking for plus or minus; a better “stopper” is an empty token, {}. (If “plus” or “minus” is there and happens to be actual text, a confusing error message will be produced, but that is rare, and beyond the scope of this discussion.)

Most packages are loaded in the preamble. There is one exception: \RequirePackage. This is usually specified before \documentclass, and is the place where certain special options should be loaded.

Some authors create a preamble that is suitable for one document, then use the same preamble for their next document, adding more packages as they go. And some unwitting newbies “adopt” such second- hand “templates” without understanding how they were created. Don’t do it!

Start with a suitable document class and add features (packages, options, and definitions) as they become necessary. Organize the loading of packages into logical groups (all fonts together, for example), and be careful not to load a package more than once; if options are needed, any loaded with a non-first \usepackage will be ignored. Some packages automatically load other packages; for example, mathtools loads amsmath and amssymb loads amsfonts. And, very important, pay attention to the order of package loading: hyperref must be loaded (almost) last; the few packages that must come after hyperref are all well documented.

Read the documentation.

Source: What every (La)TeX newbie should know