Notes on Glob's behaviour

[simonmichael]

Thank you for Glob! hledger uses it for including data files. Here are some notes I collected from studying Glob recently. Some of these should be documented more clearly, error-checked more thoroughly, and/or implemented, to improve overall usability. Related: #47.

• Glob does not expand tilde (~).
• It does not canonicalise the paths that it finds.
• The results are not in any particular order.
• The results can include directories.
• DIRPAT/ is equivalent to DIRPAT, except results will end with a double slash (//).
• A . or .. path component can match the current or parent directories (potentially including them in the results).
• * matches zero or more characters in a file or directory name.
• * at the start of a path component ignores dot-named files and directories, by default.
• **, or any number of consecutive *'s, not followed by slash is equivalent to *.
• A **/ path component matches zero or more directory parts.
• A **/ ignores dot-named directories in its starting and ending directories, by default.
• But **/ does search intermediate dot-named directories. Eg **/c can find a/.b/c. I think this is a bug, and it's hard to work around with post-processing. (See #29, reportedly fixed but not for me.)
• In zsh, with GLOB_STAR_SHORT enabled, **FOO is equivalent to **/*FOO. This means ** can match the start of file names as well, which is more convenient (and is how ** works eg in VS Code's file search). This could be a nice addition to Glob.