GameCult/Epiphany
Archived
Watch
9
0
Fork
You've already forked Epiphany
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Cut 6c mutation suite, M16-M22

Browse source 
Seven entries through the shared harness: the Superseded list checked by
length only (M16), the outcome's whole match body replaced by Ok (M17),
the fix's commit unvalidated (M18), the mutation label widened to Short
(M19), its commit widened to Short (M20), the sample estimate's line
counts transposed under a two-line anchor that carries the opener (M21),
and a claim's mutations bound widened to sixteen (M22). Each is killed by
the Cut 6c test that pins it; the key suite is blind to all seven.

The first run left M19 and M20 with no verdict: the sample and the test
spelled the label and the commit with the narrow constructors, so the
widened tree did not compile instead of admitting the forgery. Both now
go through From<&str>, which every text type has, and the forgeries
reach validation and kill.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
Meta Vens 2026-09-16 15:41:41 +02:00
commit dddf9ede53
2 changed files with 100 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

15
epiphany-pipeline/src/lib.rs
View file

@ -872,8 +872,12 @@ mod tests {
commits: vec![ReportCommit { sha: sha(), subject: "Add the pipeline documents".into(), builds: true }],
range: range(), verification: vec![evidence()],
mutations: vec![MutationRecord {
label: l("M1"), rule: "key derivation".into(), location: location(),
before: "parent_cut(field, spec, 'r')?".into(), after: "spec".into(), commit: sha(),
// `.into()` on both, not `l()` and `sha()`: the type-level
// mutations (M19, M20) widen these fields to `Short`, and
// a sample spelled with the narrow constructors would stop
// compiling instead of letting the forgery through.
label: "M1".into(), rule: "key derivation".into(), location: location(),
before: "parent_cut(field, spec, 'r')?".into(), after: "spec".into(), commit: "5f98228d".into(),
failed_as_expected: true,
}],
deviations: vec![Deviation { what: "names".into(), why: "glob exports".into() }],
@ -1671,8 +1675,11 @@ mod tests {
fn mutation_records_carry_a_dot_free_label_and_a_commit() {
let recorded = |label: &str, commit: &str| {
let mut report = report_sample();
report.mutations[0].label = Label(label.into());
report.mutations[0].commit = Sha(commit.into());
// Through `From<&str>`, which every text type has, so the
// widened types under M19 and M20 still compile and the
// forgeries reach validation.
report.mutations[0].label = label.into();
report.mutations[0].commit = commit.into();
PipelineDocument::CutReport(report).validate()
};
assert_eq!(recorded("M1", "5f98228d"), Ok(()));

89
tools/eureka-cut6c-mutations.psd1 Normal file
View file

@ -0,0 +1,89 @@
# Eureka Cut 6c mutation suite: the Ghostlight shapes. Entries only; the
# harness is tools/eureka-mutations.ps1.
#
# powershell -File tools/eureka-mutations.ps1 `
# -Entries tools/eureka-cut6c-mutations.psd1 `
# -Target epiphany-pipeline/src/lib.rs `
# -Test 'cargo test -p epiphany-pipeline --lib'
#
# A retyping is the mutation-hostile case: the compiler catches the type
# change, so the tempting test constructs a valid value and asserts `Ok(())`,
# which stays green under every mutation below. Where the cut's content is a
# type, the mutation is a type-level mutation (M19, M20, M22) and its killer
# is a forgery the wider type admits. The key suite is blind to all of these
# because a resolution's key is outcome-invariant.
@{
Mutations = @(
@{
Id = 'M16'
Rule = 'A Superseded list is validated item by item, not only bounded: a wrong-kind referent inside it is refused.'
Test = 'tests::resolution_outcome_referents_are_parsed_ids_of_their_kind'
Old = ' Self::Superseded { by } => list(&format!("{field}.by"), by, 8),'
New = ' Self::Superseded { by } => bound(&format!("{field}.by"), 8, by.len()),'
},
@{
Id = 'M17'
Rule = 'Every resolution referent is validated as a full id of the kind it declares.'
Test = 'tests::resolution_outcome_referents_are_parsed_ids_of_their_kind'
Old = @'
match self {
Self::Superseded { by } => list(&format!("{field}.by"), by, 8),
Self::Answered { by } => by.validate(&format!("{field}.by")),
Self::Fixed { commit, by } => {
commit.validate(&format!("{field}.commit"))?;
by.validate(&format!("{field}.by"))
}
Self::Deferred { to } => to.validate(&format!("{field}.to")),
Self::Recorded { reason } | Self::Withdrawn { reason } => reason.validate(&format!("{field}.reason")),
}
'@
New = @'
let _ = field;
Ok(())
'@
},
@{
Id = 'M18'
Rule = 'A fix''s commit is validated as a sha (ruling B): uppercase hex of legal length is refused.'
Test = 'tests::fixed_resolution_requires_a_commit_sha'
Old = ' commit.validate(&format!("{field}.commit"))?;'
New = ' let _ = commit;'
},
@{
Id = 'M19'
Rule = 'A mutation record''s label is a Label, so it is key-safe: a dotted label is refused.'
Test = 'tests::mutation_records_carry_a_dot_free_label_and_a_commit'
Old = 'pub struct MutationRecord { label: Label,'
New = 'pub struct MutationRecord { label: Short,'
},
@{
Id = 'M20'
Rule = 'A mutation record is pinned to a tree: its commit is a Sha, not free text.'
Test = 'tests::mutation_records_carry_a_dot_free_label_and_a_commit'
Old = 'commit: Sha, failed_as_expected: bool }'
New = 'commit: Short, failed_as_expected: bool }'
},
# The two-line anchor carries the opener: the numbers alone also match
# the report sample's StructuralDelta.
@{
Id = 'M21'
Rule = 'The estimate''s two u32 fields are not interchangeable: the sample spec estimates a net addition.'
Test = 'tests::the_sample_cut_spec_estimates_a_net_addition'
Old = @'
estimate: StructuralDelta {
lines_added: 900, lines_removed: 0,
'@
New = @'
estimate: StructuralDelta {
lines_added: 0, lines_removed: 900,
'@
},
@{
Id = 'M22'
Rule = 'A verdict claim names at most eight mutations.'
Test = 'tests::a_verdict_claim_names_the_promise_and_the_mutation_it_measured'
Old = 'mutations: Vec<Label>[8]'
New = 'mutations: Vec<Label>[16]'
}
)
}