Each typed structure in gamut_iptc::extension carries pub other: Vec<XmpProperty>: every field
the typed read took no value from, kept verbatim so a read-modify-write drops nothing. Seven types
now expose it (CreatorContactInfo, ArtworkOrObject, Licensor, ImageRegion, RegionBoundary,
RegionBoundaryPoint, Entity).
It is a public field of Vec<XmpProperty>, and that is the open question.
Against a public field
AGENTS.md asks that the public API stay mechanically portable to C: configs are "plain data —
Copy structs, fieldless enums ... or payloads reachable through accessors". A Vec of a
recursive value tree is exactly the payload an accessor pair is meant to cover.- It is writable, so a caller can put anything in it, including a name the model itself owns. The
emission rule handles that today (a retained field is emitted only when no modelled field claims
its name), but the rule exists because the collection is reachable.
For leaving it public
- Every other field of these structures is public plain data, and the types are
#[non_exhaustive], so a later move to accessors is not a breaking change for construction —
callers already build from Default and assign.
- A borrowed accessor pair (
other() / other_mut()) buys nothing over a public field while the
element type is XmpProperty, which is itself a public struct with public fields.
What to decide
- Leave
other public and document it as the retention slot (status quo).
- Replace it with
other() / push_other() / set_other() across the seven types, and say what
the accessor rejects (a field whose name the model owns? one with an empty name?).
- Something narrower — e.g. keep the field public but make the element type opaque.
Whichever is chosen should be applied to all seven types at once, and to whatever the remaining
Extension structures (#538) grow.
Raised by the review of #540, which fixed the emission rule that made a writable retention list
hazardous; the API question itself was left open there deliberately.
Refs #422
Each typed structure in
gamut_iptc::extensioncarriespub other: Vec<XmpProperty>: every fieldthe typed read took no value from, kept verbatim so a read-modify-write drops nothing. Seven types
now expose it (
CreatorContactInfo,ArtworkOrObject,Licensor,ImageRegion,RegionBoundary,RegionBoundaryPoint,Entity).It is a public field of
Vec<XmpProperty>, and that is the open question.Against a public field
AGENTS.mdasks that the public API stay mechanically portable to C: configs are "plain data —Copystructs, fieldless enums ... or payloads reachable through accessors". AVecof arecursive value tree is exactly the payload an accessor pair is meant to cover.
emission rule handles that today (a retained field is emitted only when no modelled field claims
its name), but the rule exists because the collection is reachable.
For leaving it public
#[non_exhaustive], so a later move to accessors is not a breaking change for construction —callers already build from
Defaultand assign.other()/other_mut()) buys nothing over a public field while theelement type is
XmpProperty, which is itself a public struct with public fields.What to decide
otherpublic and document it as the retention slot (status quo).other()/push_other()/set_other()across the seven types, and say whatthe accessor rejects (a field whose name the model owns? one with an empty name?).
Whichever is chosen should be applied to all seven types at once, and to whatever the remaining
Extension structures (#538) grow.
Raised by the review of #540, which fixed the emission rule that made a writable retention list
hazardous; the API question itself was left open there deliberately.
Refs #422