Bash-style brace expansion in Rust.

• Plain strings are just plain strings: abcd
• Special characters need to be escaped with a preceding \ to be considered as regular characters: ab\{cd
  • Outside of a list, special characters are {}\.
  • Inside of a list, special characters are ,{}\.
  • Inside of a sequence, special characters are ,.{}\.
• Any character preceded by a \ will be taken literally, even if it's redundant.
  • \n, for example, represents n, not a newline character.
• A List is a brace-enclosed comma-separated list of Expressions: {a,b,c}, {a,,c}, {}, {,}
  • An empty List is still a List.
  • Lists may have empty or duplicate fields.
  • Lists produce all the values of their contained Expressions.
• A Sequence is either a Numeric Sequence or a Character Sequence.
• A Numeric sequence is in the form {<spec><start>..<end>[..<stride>]}
  • <spec> is a possibly-empty set of format specifier characters:
    • = means to expand each item with leading zeroes to the width of the longest character width of <start> and <end>.
  • <start> and <end> are signed 64-bit integers. If <end> is less than <start>, the sequence will count downwards.
  • <stride> is an optional non-negative increment number, to count by increments of more than 1.
    • The default <stride> is 1.
    • A zero <stride> is always normalized to 1 to prevent infinite looping.
    • A stride may cause the endpoint to be skipped, and even the numeric size limit to be hit without error.
      • '{9223372036854775806..9223372036854775807..1000}' just produces 9223372036854775806, not an error.
• A Character sequence is in the form {<start>..<end>[..<stride>]}
  • <start> and <end> are unicode characters to produce codepoints for in order. If <end> is less than <start>, the sequence will cycle downwards. If this range would end up producing a surrogate codepoint, an error is given for each instead.
    • An error does not terminate iteration. If an error is returned, following iterations that move out of the surrogate range may still produce good values.
      • This could be used at some point to allow optional replacement characters, but I don't see a value in that over just throwing an error at this time.
  • <stride> is an optional non-negative increment number, to count by increments of more than 1.
    • The default <stride> is 1.
    • A zero <stride> is always normalized to 1 to prevent infinite looping.
    • A stride may cause the endpoint to be skipped, and even the numeric size limit to be hit without error.
      • '{a..z..1114111}' just produces a, not an error
• An Expression contains a sequence of Plain strings, Lists, and Sequences.
  • An Expression produces the cartesian product of all its items: {a,b}c{d,e}f{g..i} produces ["acdfg","acdfh","acdfi","acefg","acefh","acefi","bcdfg","bcdfh","bcdfi","bcefg","bcefh","bcefi"]
  • Expression order is produced in lexicographic order, keyed by the index of each sub-expression.
• Expressions and Lists may nest arbitrarily.
  • '{a,{b,,c{\,..\.}}{f..d..2}}' produces ["a","bf","bd","f","d","c,f","c,d","c-f","c-d","c.f","c.d"]

This does not 100% conform to Bash's style in the following ways:

• There are patterns considered ill-formed and will throw an error in bexpand. It will not try to truck along if a bad pattern is found.
• Braces are special characters and are not allowed without either being correctly formed or being escaped. In Bash a{b,c}d}e expands to abd}e acd}e and a{b{c,d}e expands to a{bce a{bde. In bexpand, both are errors.
• Empty and single-component lists are considered acceptable in bexpand. In Bash, a{}b and a{b}c are both literally repeated by the shell. In bexpand, these expand to ab and abc.
• bexpand allows character sequences to iterate any valid unicode codepoints. {🥰..🥴..2} is a valid character sequence, as is {\{..\.}, and {9..A}. Technically, {\0..\9} is valid as well, and will be treated as a character sequence, though it expands to the exact same thing as a numeric sequence of the same form. Anything that would generate an illegal unicode codepoint will generate an error.
• The width specifier is done with an equal sign at the beginning of the opening brace instead, so in Bash, {001..100} is instead done in bexpand as {=1..100}. This is to allow things like {=-5..10}, which is impossible to express in Bash.

Copyright 2023 Taylor Richberger

Published under the terms of the Mozilla Public License Version 2.0.