Two related asks that would make from.ignore (and other pattern-based fields like to.trim_prefix, to.extern, etc.) significantly more precise for large/nested C libraries like the GTK family.
1. Proper regex support, not just wildcards
Current pattern matching ('gtk_icon*', '*autoptr*', etc.) as far as I'm aware only supports glob-style * wildcards. This forces broad, blunt patterns because there's no way to express things like:
- "match
gtk_ followed by anything except source_": impossible with globs, easy with a negative lookahead or alternation in regex
- word-boundary-aware prefix matching (
gtk_tree should match gtk_tree_view_new but not gtk_treasure_thing: contrived example, but the general class of false-positive substring matches is real)
- anchoring only at start/end without wildcarding the whole pattern
Proposal: support real regex patterns, distinguished from glob patterns either by:
- a field-level flag (e.g.
from.ignore.functions.regex: true alongside a patterns: list), or
- a prefix convention (e.g.
re:^gtk_(?!source_) vs plain glob strings), similar to how many tools disambiguate glob vs regex inputs
This directly solves the GTK vs. GtkSourceView collision problem from a single ignore list instead of needing per-widget-family enumeration (gtk_icon*, gtk_font*, gtk_info*, gtk_lock*, gtk_print*, gtk_tree*, ...) or a split-config workaround. One pattern: ^gtk_(?!foo_) can replace entire manually-curated lists.
2. Recursive glob (/**) for directory includes/excludes
from.extern currently supports glob patterns but explicitly cannot glob across nested folders: the docs state "you can only glob files inside of one folder meaning that patterns like foo_library/** will not work."
For libraries with headers organized into nested subdirectories (common in larger C projects: GTK itself ships headers under multiple nested include dirs depending on distro/version), this means manually listing every subdirectory that needs to be excluded/included, which is fragile across library versions and packaging layouts.
Proposal: support standard recursive glob syntax (dir/**, dir/*/**, etc.) so a single pattern can match arbitrary depth, e.g.:
from:
extern:
- "gtk-4.0/**"instead of having to enumerate every subfolder under gtk-4.0/.
Why both together matter
These two compound well: regex for precise symbol name filtering, recursive glob for precise file path filtering. Together they'd let a rune like the GTK/GtkSourceView case collapse from an 18-entry hand-maintained ignore list plus a directory-structure-dependent extern list into a couple of expressive, version-resilient patterns, reducing maintenance burden every time a new GTK point release adds/renames a widget family or reorganizes headers.
Like the idea?
If you're on board with this expansion, I can draft a new PR adding both features, aswell as expand / correct the documentation. Let me know :)
Two related asks that would make
from.ignore(and other pattern-based fields liketo.trim_prefix,to.extern, etc.) significantly more precise for large/nested C libraries like the GTK family.1. Proper regex support, not just wildcards
Current pattern matching (
'gtk_icon*','*autoptr*', etc.) as far as I'm aware only supports glob-style*wildcards. This forces broad, blunt patterns because there's no way to express things like:gtk_followed by anything exceptsource_": impossible with globs, easy with a negative lookahead or alternation in regexgtk_treeshould matchgtk_tree_view_newbut notgtk_treasure_thing: contrived example, but the general class of false-positive substring matches is real)Proposal: support real regex patterns, distinguished from glob patterns either by:
from.ignore.functions.regex: truealongside apatterns:list), orre:^gtk_(?!source_)vs plain glob strings), similar to how many tools disambiguate glob vs regex inputsThis directly solves the GTK vs. GtkSourceView collision problem from a single ignore list instead of needing per-widget-family enumeration (
gtk_icon*,gtk_font*,gtk_info*,gtk_lock*,gtk_print*,gtk_tree*, ...) or a split-config workaround. One pattern:^gtk_(?!foo_)can replace entire manually-curated lists.2. Recursive glob (
/**) for directory includes/excludesfrom.externcurrently supports glob patterns but explicitly cannot glob across nested folders: the docs state "you can only glob files inside of one folder meaning that patterns likefoo_library/**will not work."For libraries with headers organized into nested subdirectories (common in larger C projects: GTK itself ships headers under multiple nested include dirs depending on distro/version), this means manually listing every subdirectory that needs to be excluded/included, which is fragile across library versions and packaging layouts.
Proposal: support standard recursive glob syntax (
dir/**,dir/*/**, etc.) so a single pattern can match arbitrary depth, e.g.:instead of having to enumerate every subfolder under
gtk-4.0/.Why both together matter
These two compound well: regex for precise symbol name filtering, recursive glob for precise file path filtering. Together they'd let a rune like the GTK/GtkSourceView case collapse from an 18-entry hand-maintained ignore list plus a directory-structure-dependent extern list into a couple of expressive, version-resilient patterns, reducing maintenance burden every time a new GTK point release adds/renames a widget family or reorganizes headers.
Like the idea?
If you're on board with this expansion, I can draft a new PR adding both features, aswell as expand / correct the documentation. Let me know :)