7.1.54. Glob Pattern Matching

This tutorial covers shell-style glob pattern matching in daslang. daslib ships two glob flavors that look similar but treat / differently:

daslib/fio.match_glob — pathname-aware: * stops at /, ** crosses, and [abc] / [a-z] / [!abc] character classes are supported. This is the POSIX fnmatch behavior with FNM_PATHNAME.
daslib/strings_boost.glob_match — filename-style: * matches any characters including /, ? matches one character including /, no **, no character classes. This is plain POSIX fnmatch without FNM_PATHNAME.

Use fio.match_glob for paths. Use glob_match only when matching a single name component or any text where / is incidental.

options gen2
require daslib/fio
require daslib/strings_boost

7.1.54.1. Pathname-aware matching with `match_glob`

match_glob(pattern, path) returns true when the path matches the pattern. It is pure — no filesystem access — and operates on slash-separated paths. * matches any characters except /; ** matches across /; ? matches one non-/ character; [...] is a character class.

match_glob("*.das",     "hello.das")        // true
match_glob("*.das",     "sub/hello.das")    // false — * stops at /
match_glob("**/*.das",  "sub/foo.das")      // true
match_glob("**/*.das",  "a/b/c/x.das")      // true

match_glob("[a-z][0-9].txt", "m7.txt")      // true — character class + range
match_glob("[!abc].log",     "x.log")       // true — negated class
match_glob("src/?ello.das",  "src/hello.das")  // true — ? matches one char

Empty patterns and paths follow the obvious rules: an empty pattern matches only an empty path; a path is empty-matchable only by patterns made of * / **.

7.1.54.2. Walking the filesystem with `glob`

glob(root, pattern, blk) recursively walks root and invokes blk(filename, is_dir) for every entry whose path (relative to root, normalized to use /) matches pattern. Order matches dir_rec; the same pattern works on Windows and POSIX.

var das_files : array<string>
glob("src", "**/*.das") $(filename, is_dir) {
    if (!is_dir) {
        das_files |> push_clone(filename)
    }
}

If you only need top-level entries, drop the **/:

glob("src", "*.das") $(filename, is_dir) {
    // only matches src/foo.das, not src/sub/foo.das
}

7.1.54.3. Filtering with `glob_filtered`

glob_filtered(root, includes, excludes, blk) is the multi-pattern form. An entry is yielded if it matches at least one includes pattern and zero excludes patterns — excludes win on conflict.

var includes : array<string>
var excludes : array<string>
includes |> push_clone("**/*.das")
excludes |> push_clone("tests/**")
excludes |> push_clone("_build/**")

glob_filtered("project", includes, excludes) $(filename, is_dir) {
    if (!is_dir) {
        // every .das file outside tests/ and _build/
    }
}

7.1.54.4. Detecting glob vs literal with `is_glob_pattern`

When a tool accepts a user-supplied path, it is useful to dispatch on “literal vs glob” — error if a literal file is missing, but warn if a glob matches nothing. is_glob_pattern returns true when the string contains *, ?, or [:

is_glob_pattern("main.das")      // false — literal
is_glob_pattern("*.das")         // true  — glob
is_glob_pattern("[ab]c.txt")     // true  — character class
is_glob_pattern("data/v?.json")  // true  — ?

A common convention used by build/release tools:

A literal pattern that matches no file is an error (the user named a file that does not exist).
A glob pattern that matches no file is a warning (the pattern was conditional and simply did not apply on this run).

7.1.54.5. Pattern to file list with `expand_glob`

expand_glob(pattern, var result) is the higher-level helper for the common “single glob argument from the command line” case. It splits the literal directory prefix from the glob remainder, picks recursive (glob) vs shallow (dir) walk based on whether the remainder contains ** or /, filters out directories so callers receive regular files only, sorts the matches locally, and appends them to result. Backslash patterns (src\sub\*.das) are normalized to / form internally.

var das_files : array<string>
expand_glob("src/**/*.das", das_files)
// das_files is sorted; only contains regular files; previously-existing
// entries in das_files keep their position because the sort is local.

7.1.54.6. Comma/newline lists with `parse_file_list`

parse_file_list(file, var result) is the canonical “user-supplied paths argument → expanded file list” expander used by every MCP tool with a paths / file argument. Input is a comma- or newline-separated list; whitespace is stripped; empty entries are skipped. Literal entries pass through unchanged; glob entries (those containing *, ?, or [) go through expand_glob. Order is preserved across plain entries — globs sort within their own slice.

var inputs : array<string>
parse_file_list("README.md,daslib/*.das,tests/fio/*.das", inputs)
// inputs:
//   README.md           (plain, position preserved)
//   daslib/<sorted>     (glob expansion, sorted within its slice)
//   tests/fio/<sorted>  (glob expansion, sorted within its slice)

Reach for parse_file_list from any new CLI tool that accepts a --files / --paths argument. Reach for expand_glob when the caller already split its argument into individual glob patterns.

7.1.54.7. Choosing between the two flavors

The two flavors disagree the moment / enters the picture:

let pattern = "*.das"
let path    = "sub/hello.das"

match_glob(pattern, path)   // false — * does not cross /
glob_match(pattern, path)   // true  — * matches any chars including /

Most user-facing patterns are pathname-style — release_include("assets/**/*.png"), --exclude tests/**, compile_check **/*.das — and want fio.match_glob. Reach for strings_boost.glob_match only for single-name-component matching or text where / carries no structural meaning (for example, channel-name matching).

7.1.54.8. Performance notes

match_glob accesses the pattern and path through peek_data, so byte-level matching is O(1) per character — fine inside per-file loops.
glob and glob_filtered pay the full dir_rec recursion cost. Build the file list once; do not call them per-file inside another walk.
An idiom worth knowing: peek_data("") does not invoke its block. match_glob handles this with explicit empty-string handling at the top. Any other code wrapping peek_data over potentially-empty input must do the same.

7.1.54.9. Summary

Function	Description
`match_glob(pattern, path)`	Pathname-aware glob match (true/false).
`glob(root, pattern, blk)`	Walk `root` recursively, yield matches.
`glob_filtered(root, in, ex, b)`	Walk + filter via include/exclude lists.
`is_glob_pattern(s)`	Detect whether a string is a glob or literal.
`to_generic_path(s)`	Convert a path to forward slashes regardless of host.
`expand_glob(pattern, var result)`	Pattern → sorted file list, appends.
`parse_file_list(file, var result)`	Comma/newline list of files / dirs / globs → flat list.
`glob_match(pattern, text)`	Filename-style match (`*` crosses `/`).

7.1.54. Glob Pattern Matching

7.1.54.1. Pathname-aware matching with match_glob

7.1.54.2. Walking the filesystem with glob

7.1.54.3. Filtering with glob_filtered

7.1.54.4. Detecting glob vs literal with is_glob_pattern

7.1.54.5. Pattern to file list with expand_glob

7.1.54.6. Comma/newline lists with parse_file_list