Struct AbsenceCensus 

pub struct AbsenceCensus { /* private fields */ }

How often each path was absent, counted per occurrence rather than per file.

LabelCensus counts a label once per file, which answers “does the corpus carry this at all”. It cannot answer how often a reader’s absent fallback runs, because a label inside a list is present on some elements and absent on others and a file-level count collapses that to one bit. A corpus of two hundred creatures where one omits ScriptHeartbeat and a corpus where all two hundred omit it are the same file either way.

So the denominator here is the number of structs visited at the path’s parent: the number of times the reader ran at all. A field on a list element gets one occasion per element.

Paths are label sequences with no list marker, so a schema entry looks up directly by the path a walk builds for it. A list and a struct both become the parent path of whatever hangs beneath them, which is the same nesting Shape uses for a container’s children.

Implementations

impl AbsenceCensus

pub fn new() -> Self

An empty census.

pub fn observe(&mut self, root: &GffStruct)

Folds one file’s tree into the running counts.

pub fn rate(&self, path: &[&str]) -> Option<AbsenceRate>

How often path was absent, or None when the corpus never reached its parent.

None is not the same as “always present” and not the same as “always absent”. It means no file in the corpus held a struct this field could have appeared on, so the corpus says nothing about the field either way. Ranking that as a 0% firing rate would read as evidence the fallback never runs, which is exactly backwards.

pub fn firing_rate( &self, path: &[&str], shape: SchemaShape, ) -> Option<AbsenceRate>

How often path’s fallback fires in this corpus, or None when this corpus cannot speak to it.

rate answers “how often was the label missing here”, which is not the same question. A missing label only means a fallback ran if this corpus’s reader was the one looking for it, and under SchemaShape::Union it may not have been. Callers deciding that inline have gotten it wrong in both directions:

• Treating a union schema as single-form counts every static-form label as firing across the entire saved corpus, which put six TemplateResRef rows at the head of a ranking on numbers describing a reader that never ran.
• Treating a single-form schema as a union drops every field the corpus universally omits, which is exactly the set whose fallback runs every time and therefore matters most.

So the shape is a parameter rather than a heuristic, and this is the only place the rule is written down.

pub fn occurrences(&self, path: &[&str]) -> usize

How many structs the corpus held at path.

The element count of a list, when path names one. Zero for a path the corpus never held.

Trait Implementations

impl Clone for AbsenceCensus

fn clone(&self) -> AbsenceCensus

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AbsenceCensus

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for AbsenceCensus

fn default() -> AbsenceCensus

Returns the “default value” for a type.