Any non word characters

Character classes¶

Character classes are useful to represent different groups of characters,
succinctly.

Examples:

Example:

<[^<>]+> to select an HTML tag

Shorthand character classes¶

The quantifiers¶

Greed¶

By default, with quantifiers, the regular expression engine is greedy: it
extends the selection as much as possible. This often causes surprises, at
first. ? follows a quantifier to make it lazy.
Avoid putting two in the same expression, the result can be unpredictable.

Beware of nesting quantifiers, for example, the pattern (a*)*, as it
exponentially increases processing time.

Alternation¶

The | character in a regular expression is a logical OR. It means
that either the preceding or the following expression can match.

Exclusion¶

Method 1

pattern_to_exclude(*SKIP)(*FAIL)|pattern_to_select

Example:

"Blabla"(*SKIP)(*FAIL)|Blabla

selects Blabla, in the strings Blabla or “Blabla or Blabla”, but not in “Blabla”.

Method 2

pattern_to_excludeK|(pattern_to_select)

"Blabla"K|(Blabla)

selects Blabla, in the strings Blabla or “Blabla or Blabla”, but not in “Blabla”.

Anchors¶

An anchor is a way to match a logical location in a string, rather than a
character. The most useful anchors for text processing are:

b

Designates a word boundary, i.e. a transition from space to non-space
character. For example, you can use bsurd to match the surd but
not absurd.

^

Matches the start of a line (in multi-line mode, which is the
default)

$

Matches the end of a line (in multi-line mode, which is the default)

K

Resets the start position of the selection to its position in the pattern.
Some regexp engines (but not calibre) do not allow lookbehind of variable
length, especially with quantifiers. When you can use K with these
engines, it also allows you to get rid of this limit by writing the
equivalent of a positive lookbehind of variable length.

Groups¶

(expression)

Capturing group, which stores the selection and can be recalled later
in the search or replace patterns with n, where n is the
sequence number of the capturing group (starting at 1 in reading order)

(?:expression)

Group that does not capture the selection

(?>expression)

Atomic Group: As soon as the expression is satisfied, the regexp engine
passes, and if the rest of the pattern fails, it will not backtrack to
try other combinations with the expression. Atomic groups do not
capture.

(?|expression)

Branch reset group: the branches of the alternations included in the
expression share the same group numbers

(?<name>expression)

Group named “name”. The selection can be recalled later in the search
pattern by (?P=name) and in the replace by g<name>. Two
different groups can use the same name.

Lookarounds¶

Lookaheads and lookbehinds do not consume characters, they are zero length and
do not capture. They are atomic groups: as soon as the assertion is satisfied,
the regexp engine passes, and if the rest of the pattern fails, it will not
backtrack inside the lookaround to try other combinations.

When looking for multiple matches in a string, at the starting position of each
match attempt, a lookbehind can inspect the characters before the current
position. Therefore, on the string 123, the pattern (?<=d)d (a digit preceded
by a digit) should, in theory, select 2 and 3. On the other hand, dKd can
only select 2, because the starting position after the first selection is
immediately before 3, and there are not enough digits for a second match.
Similarly, d(d) only captures 2. In calibre’s regexp engine practice, the
positive lookbehind behaves in the same way, and selects only 2, contrary to
theory.

Groups can be placed inside lookarounds, but capture is rarely useful.
Nevertheless, if it is useful, it will be necessary to be very careful in the
use of a quantifier in a lookbehind: the greed associated with the absence of
backtracking can give a surprising capture. For this reason, use K rather than
a positive lookbehind when you have a quantifier (or worse, several) in a
capturing group of the positive lookbehind.

Example of negative lookahead:

(?![^<>{}]*[>}])

Placed at the end of the pattern prevents to select within a tag or a style embedded in the file.

Whenever possible, it is always better to “anchor” the lookarounds, to reduce
the number of steps necessary to obtain the result.

Recursion¶

Recursion is calling oneself. This is useful for balanced queries, such as
quoted strings, which can contain embedded quoted strings. Thus, if during the
processing of a string between double quotation marks, we encounter the
beginning of a new string between double quotation marks, well we know how to
do, and we call ourselves. Then we have a pattern like:

start-pattern(?>atomic sub-pattern|(?R))*end-pattern

To select a string between double quotation marks without stopping on an embedded string:

“((?>[^“”]+|(?R))*[^“”]+)”

This template can also be used to modify pairs of tags that can be
embedded, such as <div> tags.

Special characters¶

Meta-characters¶

Meta-characters are those that have a special meaning for the regexp engine. Of
these, twelve must be preceded by an escape character, the backslash (), to
lose their special meaning and become a regular character again:

Seven other meta-characters do not need to be preceded by a backslash (but can
be without any other consequence):

Special characters lose their status if they are used inside a class (between
brackets []). The closing bracket and the dash have a special status in a
class. Outside the class, the dash is a simple literal, the closing bracket
remains a meta-character.

The slash (/) and the number sign (or hash character) (#) are not
meta-characters, they don’t need to be escaped.

In some tools, like regex101.com with the Python engine, double quotes have the
special status of separator, and must be escaped, or the options changed. This
is not the case in the editor of calibre.

Modes¶

(?s)

Causes the dot (.) to match newline characters as well

(?m)

Makes the ^ and $ anchors match the start and end of lines
instead of the start and end of the entire string.