Re-working <https://github.com/armedbear/abcl/pull/743> from <https://github.com/blakemcbride/abcl/tree/conditions-packages-types-structs>#748
Open
easye wants to merge 11 commits intoarmedbear:masterfrom
Conversation
4 tasks
TODO: reword commit, at least noting fasl incompatiblity at the top.
- Fixes all 12 failing `MAKE-LOAD-FORM.ORDER.*` tests (ORDER.3–14) in
the ANSI test suite: compiled-test failures drop from 64 to 52
with no regressions.
- The file compiler now emits `MAKE-LOAD-FORM` creation and
initialization forms to the fasl in data-flow dependency order,
with init forms run ASAP per CLHS.
- Bumps `*FASL-VERSION*` from 43 to 44 (fasl layout changed).
CLHS requires, for every literal instance handled by
`MAKE-LOAD-FORM`:
- the creation form runs after creation of every object it
references (transitive closure);
- the init form runs after all objects it references are created,
and as soon as its data-flow dependencies are satisfied.
Previously, ABCL emitted the creation + init forms inline at the
point where the instance was referenced, expanded via
`get-instance-form` as `(let ((inst <creation>)) <init> inst)`. This
could not express cross-object dependencies and failed every
`MAKE-LOAD-FORM.ORDER.*` test that involved more than one instance
with dependencies between them.
Each fasl now carries a per-file instance vector
`sys::*fasl-instances*` indexed by integer. For every literal
instance referenced in the compile unit, the compiler emits (in
dependency order, *before* the top-level form that references it):
(setf (svref sys::*fasl-instances* N) <creation-form>)
<initialization-form>
and the original inline reference to the instance is rewritten to
`#.(svref sys::*fasl-instances* N)`. The vector is allocated once in
the fasl prologue from the final instance count.
A two-phase dependency walker drives the emission:
1. `%fasl-ensure-created(X)` — recursively ensures creation of every
instance embedded in `X`'s creation form, then emits `X`'s
creation form. A re-entry while `X` is mid-creation is a real
circular creation dependency and signals an error (CLHS: undefined
behavior).
2. `%fasl-ensure-initialized(X)` — ensures `X` is created, then
recursively ensures initialization of every instance embedded in
`X`'s init form, then emits `X`'s init form. An in-progress init
cycle is tolerated (CLHS leaves ordering within an init cycle
unspecified).
To honor the ASAP rule (tests `.6`, `.10`–`.12`),
`%fasl-ensure-created` eagerly initializes `X` immediately after
emitting its creation form *iff* `%fasl-init-deps-ready-p` reports
that every object referenced by `X`'s init form (ignoring `X` itself,
since MAKE-LOAD-FORM methods commonly embed `',x` in their init
form) is already initialized. Otherwise init-`X` is emitted later,
either from the top-level walk that directly references `X` or from
another object's init-form walk.
`dump-instance` now writes `#.(svref sys::*fasl-instances* N)`
instead of the legacy inline form. The existing `*circularity*`
walker (`df-check-object`) had to be taught to walk *the same* cons
that `dump-instance` ultimately writes, so a single reference cons
per instance is cached in `*fasl-instance-refs*` and reused by both
sites.
A new dynamic variable `*fasl-emitting-to-fasl-stream*` gates the new
behavior: when unbound (e.g. dumping constants into a class file via
`SERIALIZE-OBJECT`), `dump-instance` / `df-check-instance` fall back
to the legacy `get-instance-form` expansion.
TODO: test, understand and write a better summary comment CLHS: the :predicate option is limited to use with structures that have no :type option or are typed and :named. A non-default (user-supplied) predicate name is a symbol or NIL; the default is a string placeholder. CLHS: if :type is supplied and :named is supplied, the type must be able to hold the structure's name (a symbol).
TODO: verify this is what we want FuncallableStandardObject.typep() previously returned T for COMPILED-FUNCTION` when the installed dispatcher happened to be compiled; this is an implementation detail of the funcallable instance, not a property of the object itself. Returning NIL matches SBCL/CCL behavior. Because `jvm-compile` used the old T answer as an early-return signal, it now explicitly skips `funcallable-standard-object` definitions (they have no LAMBDA expression to work from). A funcallable-standard-object (e.g. a generic function) is not itself a COMPILED-FUNCTION regardless of whether its installed dispatcher function happens to be compiled: COMPILED-FUNCTION characterizes the object, not an internal implementation detail. Returning T here breaks TYPE-OF invariants (SUBTYPEP of the object's type would have to cover COMPILED-FUNCTION).
easye
force-pushed
the
conditions-packages-types-structs
branch
from
367a7dc to
a008e76
Compare
added 7 commits
April 23, 2026 10:19
- **`STANDARD-GENERIC-FUNCTION` ↔ `COMPILED-FUNCTION`.** `FuncallableStandardObject.typep` previously returned T for `COMPILED-FUNCTION` when the installed dispatcher happened to be compiled; this is an implementation detail of the funcallable instance, not a property of the object itself. Returning NIL matches SBCL/CCL behavior. Because `jvm-compile` used the old T answer as an early-return signal, it now explicitly skips `funcallable-standard-object` definitions (they have no LAMBDA expression to work from).
TODO: verify
`TYPE-OF.1` / `.4` require that for every object `x`, `(type-of x)` be
a subtype of every built-in type that `x` is `TYPEP`. Two specific
instances failed this invariant in ABCL:
A
(simple-array nil (*))
reported
(typep x 'base-string) → T
but
(subtypep '(nil-vector n) 'base-string) → NIL, T
because the upgraded element types (`nil` vs `character`) are disjoint
per CLHS ` 4.2.3. `NilVector.typep` now returns NIL for BASE-STRING
and SIMPLE-BASE-STRING. STRING remains T — NIL is a subtype of
character, so a NIL-vector is legitimately a string.
TODO: verify, reword - **`subtypep` bugfix in the string-array branch** of `csubtypep-array`: when the second type was a bare symbol like `nil-vector` (no dimension spec), the size check compared `(car i1)` against `nil` instead of treating the missing dimension as `*`. `(subtypep '(nil-vector 0) 'nil-vector)` now returns `T, T`.
TODO: split off CALL-NEXT-METHOD fixes Three independent bugs in the long-form `define-method-combination` expansion: - **`:order` / `:required` evaluated as forms, not quoted symbols.** `canonicalize-method-group-spec` was emitting `:order ',order` and `:required ',required-p`, so a group spec like `((method-list * :order order))` substituted the literal symbol `order` rather than the value of the lexical variable. CLHS specifies these are *forms* to be evaluated at call site. - **Optional parameters "in excess" of the generic function's optional count were bound to NIL** instead of their initforms. Rewrote the optional-parameter handling in `method-combination-type-lambda-with-args-emf` to emit an initform binding for excess optionals, and to bind excess supplied-p parameters to NIL. - **Keyword `supplied-p` parameters leaked the `member` tail.** The same binding symbol was used for both the boolean supplied-p and the `(cadr …)` accessor, so user code saw e.g. `(:Z1 4)` instead of `T`. Split these via a private gensym for the `member` result; `supplied-p` now binds to `(not (null …))`. AMOP requires that when CALL-NEXT-METHOD is passed an argument list, those arguments must produce the same set of applicable methods as the original call. ABCL previously silently re-dispatched on whatever arguments it received. Added `*call-next-method-gf*` / `*call-next-method-applicable-methods*` specials bound by the slow-method-lookup path, plus a `cnm-check-applicable-methods` helper invoked from the `flet-call-next-method` expansion when explicit CNM args are given. The emfun is wrapped so the binding survives effective-method caching. TODO --- SPLIT? AMOP requires that when CALL-NEXT-METHOD is passed an argument list, those arguments must produce the same set of applicable methods as the original call. ABCL previously silently re-dispatched on whatever arguments it received. Added `*call-next-method-gf*` / `*call-next-method-applicable-methods*` specials bound by the slow-method-lookup path, plus a `cnm-check-applicable-methods` helper invoked from the `flet-call-next-method` expansion when explicit CNM args are given. The emfun is wrapped so the binding survives effective-method caching.
CLHS specifies that multiple :nicknames options are to be merged.
DEFPACKAGE was using `setq`, so only the *last* `:nicknames` clause
survived:
(defpackage :foo (:nicknames :a) (:nicknames :b)) ; only :B seen
TODO: Find correct CLHS citation as §2.4.8.7 <http://clhs.lisp.se/Body/02_dhg.htm> does not require signalling a reader error as noted below. CLHS 2.4.8.7 requires `#:<token>` to signal a reader error if the token contains an unescaped package marker (a colon not preceded by `\` or inside a `|…|` escape). ABCL previously accepted `#:foo:bar` silently and constructed a symbol whose name contained a colon. `Stream.readSymbol` now scans the token post-`_readToken`, consulting the per-character escape `BitSet` returned by the tokenizer, and signals `ReaderError` on any unescaped `:`. The check is skipped when `*read-suppress*` is true, so `READ-SUPPRESS.SHARP-COLON.7` continues to pass (it reads `"#::"` with `*read-suppress*` = T and expects `NIL`, no error).
Addresses problems with the ANSI-TEST DISASSEMBLE.ERROR.3 expecting TYPE-ERROR for invalid arguments. Note: an improved version of this sort of checking would make the effort to see if a given symbol was defined in the function space, if a given SETF expander had a definition, etc. SBCL seems to do this.
easye
force-pushed
the
conditions-packages-types-structs
branch
5 times, most recently
from
63560af to
ceb4b36
Compare
CLHS defines APROPOS and APROPOS-LIST with the lambda-list
(string-designator &optional package-designator)
ABCL extends ANSI by adding a third EXTERNAL-ONLY parameter to list
only external symbols.
(string-designator &optional package-designator external-only)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
IN-PROGESS
#:, APROPOS, and DISASSEMBLE ANSI test failures #743 into logical chunks