One bracket, one position

A bracket constrains exactly one character position

A bracket expression matches one character of a filename, chosen from a set you write out. [abc] matches an a, a b, or a c — one of them, never two, never zero. Everything difficult about brackets comes from forgetting that sentence.

Inside the brackets you may write:

• literal characters, in any order: [cbA] and [Abc] are the same set.
• ranges, written with a hyphen: [0-9], [a-z], [A-Z]. A range means "every character whose code lies between these two, inclusive", using the machine's character codes — which is why [a-Z] is not a range at all.
• negation, written as [!…] (the POSIX spelling) or [^…] (the widely-supported alias the textbook uses). [!5] matches any single character other than 5.

You can combine them: [0-9a-zA-Z] is one position that must be alphanumeric, and [!0-9a-zA-Z] is one position that must not be.

Worked example — two patterns, one directory

The directory holds six files:

lab1.c  lab1a.c  lab2b.c  lab5c.c  lab12x.c  labxy.c

Pattern one: lab[0-9]?.c. Take it one candidate at a time. The pattern demands the literal letters l, a, b, then one digit, then any one character, then a literal .c — a name of exactly eight characters.

• lab1.c — after lab comes 1, good; the ? must then consume ., and the pattern still needs .c but only c is left. No match.
• lab1a.c — 1 is a digit, a fills the ?, .c matches. Match.
• lab2b.c — same shape. Match.
• lab5c.c — same shape; the 5 is a perfectly ordinary digit here. Match.
• lab12x.c — 1 is the digit, 2 fills the ?, and then the pattern wants . but finds x. No match.
• labxy.c — x is not a digit. No match.

Pattern two: lab[!5]*.c. Now the fourth character must be anything except 5, and the asterisk absorbs whatever is left before .c.

• lab1.c — fourth character 1, not a 5; the asterisk matches zero characters. Match.
• lab1a.c, lab2b.c, lab12x.c, labxy.c — all match for the same reason.
• lab5c.c — fourth character is 5. No match, and it is the only one excluded.

Notice that the negation excluded exactly one position, not a substring: lab12x.c contains no 5 at all, but that is irrelevant; the pattern only ever looked at character four.

The three misreadings that cost marks

A comma inside brackets is a literal comma. [6,7] is a set of three characters: 6, , and 7. It matches a filename position holding a comma just as happily as one holding a six. This trips people because the textbook itself writes [c,C] as though the comma meant "or" — it works there only because no filename in the example happens to contain a comma.

A range is a range of character codes, and letters are not contiguous. In ASCII the uppercase letters occupy codes 65–90 and the lowercase letters 97–122, with six punctuation characters in the gap. So [aA-Z] is not "a or A through Z" in the sense of "any letter" — it parses as the literal a followed by the range A–Z, giving twenty-seven characters: every uppercase letter, plus lowercase a, and no other lowercase letter. Writing [a-Z] instead is worse: it asks for a range that runs backwards, which the shell either rejects or silently treats as empty.

A bracket set is not a word. [comp] is one position that must hold c, o, m or p. It does not match the four letters comp; a pattern that did would be written comp, with no brackets at all.

Reading a whole assignment pattern

Put those together on rm *[aA-Z]??[6,7].[^j]*, which is exactly the shape an assignment asks about. Left to right: any run of characters, then one character that is a or an uppercase letter, then any two characters, then one character that is 6, , or 7, then a literal dot, then one character that is not j, then any run of characters. Note the last two: [^j] constrains only the first character after the dot. A file called report,x.json — dot, then j — survives; a file called report,x.txt does not.