You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Generated by IronKernel.Tests/KernelConformanceTests.fs. Do not edit by hand;
run the suite with IRONKERNEL_UPDATE_CONFORMANCE=1 to regenerate.
Features are those of the [Revised-1 Report on the Kernel Programming Language]
(https://ftp.cs.wpi.edu/pub/techreports/pdf/05-07.pdf) (WPI TR 05-07), extracted
from its table of contents by tools/extract-r-1rk-features.py into
docs/kernel-r-1rk-features.json. The report decides what must be covered, so
features IronKernel has never implemented still appear here.
Status
Meaning
verified
Every binding resolves and a behavioural check for this entry passes. The check exercises the entry; it does not prove every requirement in it.
bound
Every binding resolves, but nothing checks its behaviour yet. This is not evidence of conformance.
partial
Some bindings of the entry resolve and others do not.
absent
No binding of the entry resolves.
IronKernel is a dialect, and drops the $ sigil the report uses for operatives
($if is spelled if, $define! is spelled define). Where a feature resolves
under such a name the matrix records it as an alias, so the divergence is visible
rather than hidden behind a pass.
Summary
Entries
Share
verified
135
100%
bound
0
0%
partial
0
0%
absent
0
0%
total
135
34 of 135 entries belong to modules the report marks optional; an
implementation may omit them and still conform.
Known divergences
Differences a status cannot express. A verified row means the entry was
exercised, not that IronKernel matches the report exactly.
Report
Divergence
7.2.5
Two things. First, applying a continuation object directly, as in (k 1), bypasses the selection and interception of 7.2.5: no guard is consulted. That is a dialect extension rather than a gap in the report, which does not make continuations applicable at all -- apply-continuation (7.3.1) and continuation->applicative (7.2.5) are its mechanisms, and both intercept correctly. Anything relying on a guard should use those, and since 7.2.7 now signals by abnormal pass, errors do go through interception while a direct application still does not. Second, a combination's operand tree is represented as a list, so the atomic operand tree the report also allows has no spelling: (divert #f) delivers (#f) rather than #f, which is why the report's $binds? derivation (6.7.1) needs a list on both paths here. apply-continuation takes the object directly and is unaffected.
3.6
Two spellings differ from the report's, and both round-trip through read, because IronKernel's reader accepts them. A dotted pair writes as (1 & 2) rather than (1 . 2), since . is the CLR interop operative here. A CLR value with no Kernel representation -- a DateTime, a complex number, an inexact infinity or NaN -- writes as <obj v : T>, which the reader cannot parse; 3.6 provides for objects that have no external representation, and those are IronKernel's.
12.9
The report specifies 12.9.2 through 12.9.6 by signature only. Appendix A.2 records that it is an incomplete draft whose unwritten portions were "only planned in rough outline", so verified there means the binding exists with its standard mathematical meaning, and the choices the report leaves open are IronKernel's: an argument outside a function's real domain takes its complex value now that 12.10 is supported, so (sqrt -1) is i; a result with no primary value follows 12.2's rule and so signals under strict arithmetic and is returned without it; and infinities are returned as values.
12.2
Inexact reals are non-robust with bounds of exact negative and positive infinity, which 12.2 sanctions explicitly ("an implementation can fully support module Inexact without making any effort to maintain finite bounds or robustness"). Two things follow. robust? is exactly "every argument is exact". And no number is ever created with its lower bound above its upper bound, so the report's undefined number never arises and undefined? is always false. Narrowing the bounds is what module Narrow inexact (12.7) asks for, and that module is absent.
12.7
Module Narrow inexact is supported in the sense the report requires -- the variable binds and reads, and the bounding information is no less restrictive when it is set than when it is cleared -- but setting it changes nothing observable. Narrowing is advice ("the implementation is advised to maintain the most restrictive bounding and robustness information it (correctly) can"), and IronKernel maintains the infinite bounds of 12.2 either way.
12.3.3
Under strict arithmetic, overflow and a result with no primary value both signal; underflow does not, and still gives zero. Telling an underflow apart needs the operation and not just its operands -- a zero result from non-zero operands is an underflow for multiplication and an exact answer for subtraction -- so it is left undetected rather than guessed at. The variable's initial value here is true, which the report leaves open.
7.2.6
root-continuation is not literally at the end of every continuation chain: IronKernel's drivers give each top-level form its own continuation. Its extent is instead defined to contain everything, which is what "the ancestor of all other continuations" means for the selection algorithm, so a clause selecting on it is always selected. Receiving a value ends the session, and the process exit status is 0.
4.7
Module Pair mutation is complete. Mutability follows where the structure came from: the reader produces immutable pairs, because a program is an algorithm rather than data the program made, while cons and list produce mutable ones. R-1RK 6.4.1 makes it an error for two arguments of append! to share a last pair; that condition is on the caller and is not checked, so appending such a pair to itself builds a cycle rather than signalling.
6.3
The derived list library divides on whether a derivation asks about a list's shape or walks its elements. length, finite-list? and countable-list? go through get-list-metrics, which measures a cycle rather than walking into one. list-tail, filter, reduce, append and the rest walk elements and diverge on a cyclic argument, as the report's own derivations of them do. reduce (6.3.10) is the exception and is complete: its six-argument syntax reduces a cyclic list through the caller's precycle/incycle/postcycle, with the call counts the report fixes. Its three-argument syntax still diverges on a cycle, which is what the report specifies -- there the second syntax "must be used".
12.10
Complex numbers are System.Numerics.Complex, so components are double precision and a result whose imaginary part is zero collapses back to a real. The report specifies 12.10 by signature only.
12.8
Exactness is carried by a value's representation rather than by a separate tag: 1/2 is an exact ratio and 0.5 is a double, and exact? reads the representation. Module Inexact (12.6) is supported on that basis. One consequence is that numerator and denominator of an inexact real return exact integers -- (denominator 0.1) is 2^55 -- rather than inexact ones.
Modules
R-1RK 1.3.2 makes the module the unit of conformance: "An implementation
cannot claim to support a module M unless it both (1) supports all of the
features in M, and (2) supports all of the modules assumed by M."
"All entries verified" is weaker than "supported". The column below says
only that every entry of the module has a passing behavioural check. It does
not assert 1.3.2 support, which additionally requires the module's assumed
modules and the report's baseline representation requirements. Read the
divergences before taking any row as a claim of conformance.
Module
Required
Entries
Verified
All entries verified
4.1 Core types and primitive features — Booleans
required
1
1
yes
4.2 Core types and primitive features — Equivalence under mutation (optional)
optional
1
1
yes
4.3 Core types and primitive features — Equivalence up to mutation
required
1
1
yes
4.4 Core types and primitive features — Symbols
required
1
1
yes
4.5 Core types and primitive features — Control
required
2
2
yes
4.6 Core types and primitive features — Pairs and lists
required
3
3
yes
4.7 Core types and primitive features — Pair mutation (optional)
optional
2
2
yes
4.8 Core types and primitive features — Environments
required
4
4
yes
4.9 Core types and primitive features — Environment mutation (optional)
optional
1
1
yes
4.10 Core types and primitive features — Combiners
required
5
5
yes
5.1 Core library features (I) — Control
required
1
1
yes
5.2 Core library features (I) — Pairs and lists
required
2
2
yes
5.3 Core library features (I) — Combiners
required
2
2
yes
5.4 Core library features (I) — Pairs and lists
required
1
1
yes
5.5 Core library features (I) — Combiners
required
1
1
yes
5.6 Core library features (I) — Control
required
1
1
yes
5.7 Core library features (I) — Pairs and lists
required
2
2
yes
5.8 Core library features (I) — Pair mutation (optional)
optional
1
1
yes
5.9 Core library features (I) — Combiners
required
1
1
yes
5.10 Core library features (I) — Environments
required
1
1
yes
6.1 Core library features (II) — Booleans
required
5
5
yes
6.2 Core library features (II) — Combiners
required
1
1
yes
6.3 Core library features (II) — Pairs and lists
required
10
10
yes
6.4 Core library features (II) — Pair mutation (optional)
optional
4
4
yes
6.5 Core library features (II) — Equivalence under mutation (optional)
optional
1
1
yes
6.6 Core library features (II) — Equivalence up to mutation
required
1
1
yes
6.7 Core library features (II) — Environments
required
10
10
yes
6.8 Core library features (II) — Environment mutation (optional)