DISCUSS.org's sentinel-fill idea, built as two builtins because the author
asked for both: a memset with a byte the program picks, and the fixed
DE AD BE EF pattern a hex dump reads as DEADBEEF.
Both are spelled the way (zeroed) is — the value of whatever type is
expected of them — so (set grid (filled 0xFF)) fills a place and there is
no second, place-taking form beside set.
What may be filled is numbers, and structs and fixed arrays built out of
them. Everything else is refused by name: a filled dyn is a collector root
pointing at nothing, a filled Vec header frees a wild address, a filled
slice length is a bounds check that passes, and a filled bool is an i1 to
LLVM and a whole byte to x86, which is the one divergence this feature
cannot have.
The byte fill is llvm.memset / rep stosb. The four-byte pattern cannot be
a memset on either side — the intrinsic takes one repeated i8 — so it is a
counted dword loop in emit.ml and rep stosd in x86.ml, with the pattern
bytes and their little-endian word living once, in Emit. A size that is
not a multiple of four ends on DE, DE AD, or DE AD BE.
A bare {.field v} had its refusal in Parse.expr, before any checking, so a
defn whose return type was the only place the struct's name appeared could
not build one. The refusal moves to Check: Parse builds an Ast.Bare out of
the same struct_fields the named form uses, and check_bare reads the type
name off the expectation and hands that very list to check_struct. ZII, the
unknown-field refusal and the duplicate-field refusal are therefore not
copies of the named form's rules but the named form's rules.
Braces at a dyn want are the dyn map literal and stay exactly that. A
.field-keyed brace was never part of that spelling, and at a dyn want it is
refused by name rather than given a second meaning.
(Cell 1 2) is the other half, and it is character-for-character an ordinary
call, so only the symbol table separates them. It is decided on the last arm
of named_call, after a local of function type, a generic and the function
table -- so a defclass constructor, which is a real defn, resolves above it
and is untouched. Arity is exact: ZII is what the braces do, and a positional
list cannot say which field it left out, so it is not allowed to leave one
out. The refusal names the first field it did not reach and points at the
spelling that does mean "zero the rest".
Both are gone before any backend sees them -- Tast.Make either way -- and the
three acceptance rows print the same lines to say so.
A let binds in sequence, so binding a method's names pairwise from the
generic's reads a name it has just bound. A generic [a b] with a method
[b a] -- a swap, which is what renaming parameters most often is -- was
handed its first argument twice and could not reach its second at all;
[b c] is the same bug one step shorter. Every argument is now copied
into a temp in the unspellable ~ namespace first and every method name
bound from a temp, uniformly rather than only for the pairs that
collide, because a rule that fires on the tangled case alone is one
nobody exercises. Both shapes are in dyn-class.flan, where the values
are what is wrong rather than the types, and across all three rows.
With it, two things the descriptor fix left behind. descriptors_asm
wrote the descriptors into .rodata and a descriptor holds the address of
its own offset table, so every one of them was a relocation in a
read-only section -- a DT_TEXTREL, which ld warns about in a PIE and
refuses in a shared object, and which was warning in the new daemon
case's own output. They go in .data.rel.ro now, in both the executable
and the reload module; readelf -d on a reload module from each backend
shows no TEXTREL. And FIX.org: the stale held line for item 6, the
fourth read site of the shape tag (say_render, not just print), the
warning that a class's qualifier is the importer's alias so a
hand-written :a/point is coupled to one import's name, and the gap
flagged for the next sweep -- marking through a descriptor an x86 reload
module emitted is still unexercised.
A defclass is a named dyn map with a shape tag, and a generic function
dispatches on it two ways: CLOS's, where the dispatch value is the class
of the first argument, and Clojure's, where a body computes it. They are
one mechanism and not two — a class dispatcher is (class-of arg0) as the
dispatch function, which is what lets a method written for the class
point and one written for the value :point be the same branch.
(defclass point [x y])
(point 3 4) ; the constructor, positional
(class-of p) ; :point, or nil for anything else
(defgeneric area [self] dyn)
(defmethod area point [p] (* (get p :x) (get p :y)))
(defmulti describe [x] dyn (get x :kind))
(defmethod describe :square [s] ...)
(defmethod describe :else [s] ...)
A slot is a key in the instance's own map, so get, put and has-key? are
how one is read and written and no operation was added for any of it.
What the class adds is the tag, and the tag lives in the object's header
rather than in a reserved entry — the queue's note said a reserved key
and this departs from it, because a key would be counted by len, walked
by the renderer and compared by equality, so every instance would answer
a length one larger than its slot count and print a key nobody wrote. A
header field cannot be reached by get or put at all, so no user key can
collide with it. It costs nothing: the map arm of flan_obj's union grows
to the size the view arm already had, and sizeof(flan_obj) is unchanged.
It needs no tracing either — the tag is an interned keyword entry, which
is immortal and is not a collector object.
The tag shows up in exactly three places: class-of answers it, equality
compares it (two instances of one class compare by their slots; an
instance and a plain map with the same entries do not, which is
Clojure's answer for a record beside a map), and both renderers print it
— #point{ :x 1 :y 2}, Clojure's own spelling.
None of the four forms reaches the checker. lib/classes.ml turns the
whole declaration list into ordinary defns at the top of build_program,
the way Shim.expand already turns a declare-c into a declare plus a
defn: a class becomes its constructor, a generic becomes one function
whose body binds the dispatch value and compares it down a chain, and a
method becomes a branch of that chain. It is a pass and not a macro
because a macro sees one form and the generic's body is not decidable
until every method is in hand — a method may be written above its
generic, below it, or arrive at a reload an hour later.
That last case is why the method bodies are inlined rather than lifted.
A generic is exactly one top-level name, so adding a method to a running
program is the ordinary redefinition of one function, through the cell
every call site already goes through. session.ml names the generic
alongside the method's own declaration name for that reason. The cost,
recorded rather than hidden: a method is not separately callable and is
not a frame of its own.
A dispatch that finds no method signals NoMethod, a prelude struct
carrying the generic's name and the dispatch value that missed. A
condition and not a trap, because a miss is something a program can be
written to answer, and handler-case around the call is the shape. Its
value field is dyn, the first condition here with one; the per-type
descriptor an item-2 struct carries is what the collector reaches it by.
No restart is established at the miss, which is BoundsError's decision
taken for BoundsError's reason.
Both backends, identically: the two new runtime entry points are
declared in emit.ml and the x86 backend needs nothing, since a dyn call
is a dyn call there. Deferred and written down in FIX.org: inheritance,
multi-argument dispatch, :before/:after/:around, named-slot
construction, unknown-slot checking, and computed dispatch values.
The refusal moves to the checker: Emit.const refused a computed defconst by
name while the x86 backend ran it through the startup function behind an
.init~once. flag, like a defvar, so the two backends disagreed about the same
program. One refusal in Check.const_defconst_init ends that, and it is the only
place that can name the way through.
The accepted set is unchanged: Tast.const_init's, which is Emit.const's and the
x86 data_sym path's, plus the integer arithmetic collect's folding pass has
already turned into an Int before the initialiser is looked at.
Emit.const's two refusals become a failwith no program reaches; emit_global's
gconst || const_init loses its left half; x86 needed no edit, since it never
classified by the form. test_flan's infers probe asks Check.expression now that
a defconst can no longer wrap an arbitrary expression.
Ten test binaries share a directory and had shared nothing in it but
watchdog.ml. Everything else each one needed it wrote out again: the
failure counter and its FAIL line, the three-line report tail, a poll,
a socket connect, the wait for a [flan dev] daemon to bind, a substring
search, and the Load -> Check -> Reach.link front half of a compile.
[listening] was the clearest case. Three copies, byte for byte apart
from one comment, and two of them said in that comment that they were
kept separate because "these three files have no module between them".
That was not true when it was written: watchdog.ml was already named in
the same (modules ...) stanzas. test_support.ml is the second such
module, wired the same way, and those two sentences go with the copies
they were explaining.
test_repl.ml's [quote] was Wire.quote character for character, in a file
that already links Wire and already names Wire.quote in a comment about
what the case below it is checking. It is Wire.quote now.
One real behaviour change, and it is a fix. [connect] existed twice over
with different retries: the agent's narrowed to ECONNREFUSED with a
comment saying why -- the socket file appears at bind, a moment before
listen -- while dev's and repl's retried any Unix_error, which meant an
ENOENT or an EACCES was retried to the full timeout before raising
something the reader still had to interpret. The shared one takes the
narrow version. Every caller connects to a socket [listening] has
already seen on disk, so the race it does catch is the only one left.
The rest is left where it is, on purpose. The three output-capturing
[run]s differ in what they wrap -- a pid suffix, a sanitizer environment,
a valgrind invocation -- and are not the same function. The report tails
in test_repl, test_web and the two sweep binaries print different things
for different reasons. The per-file scratch prefixes are the feature that
keeps two suites running at once from unlinking each other's sockets, so
the shared helper takes the prefix rather than choosing one. And the
[match Sys.command "command -v clang ..."] probes stay as they are:
their skip lines are output this suite pins.
bin/main.ml has the compile pipeline written out twice more. Left alone
-- this was a test/-scoped change and bin/ should not be reaching into a
test module -- and noted in FIX.org as what it actually needs, which is
the pipeline moving into lib/.
dune test: exit 0, and its output is the same line for line once the
temp-directory hash and the millisecond counts are normalised.
There is no SSE remainder instruction, and LLVM does not invent one: at -O0
it lowers frem to fmod or fmodf. The backend now calls those two symbols
rather than refusing the operator, which is agreement by construction rather
than a second hand-written identity that would have to get every rounding,
every signed zero and every infinity right on its own.
Rem was the only gap. emit.ml's float surface is Add, Sub, Mul, Div, Rem and
the six comparisons; x86 had everything but Rem, and its comparisons already
build LLVM's ordered predicates out of ucomis, setcc and setnp.
math3.flan grows the operator spelling beside the fmod-f32/fmod-f64 calls it
already had, through globals so the pair is not folded before either backend
sees an operator. FIX.org records the ruling the fix came from.
(and a b) desugared to (if a b false), so it answered the last operand
only when every operand was truthy; a falsey one came back as a bare
false, where Clojure answers the falsey operand itself. It now uses the
same expansion or got in ad0f1fb -- (let [t a] (if t b t)) against or's
(let [t a] (if t t b)) -- so the operand that decided the form is the
answer, and the test is still evaluated exactly once.
The temp binding and its if now carry the operand's own loc rather than
the whole form's, which the or fix had lost: (or (vec-new i32) v) blamed
the enclosing form at 3:13 and now points at 3:18, the operand, and and's
second operand gained the same precision.
The parse pins in test_flan.ml now tie the bound name to the temp the if
tests and the bound value to the first operand, so a desugaring that
dropped the temp and wrote the operand into the arm twice no longer
passes; and has its own pin. dyn-if-truthy.flan grows the falsey-nil and
falsey-false answers, 0 and "" as truthy operands, one- and zero-operand
forms, and a printing operand that proves both the short circuit and the
single evaluation.
One behaviour that used to compile changed: with both arms of the
desugared if now holding real values, (and dyn-value typed-bool) unifies
on the typed arm and a non-bool dyn decider traps at the strict bool
boundary -- (and (box nil) some-bool) printed false and now traps, the
mirror of what (or false (box "s")) already did on dev-loop. Recorded in
FIX.org as the author's call on how check_if should join a bool arm and a
dyn arm.
The [At] arm of [permanent_root] recursed through any indexed target, so an
element of a global SLICE answered permanent the way an element of a global
ARRAY does. An array's elements are inside the global's storage; a slice's
are ptr+len pointing wherever, which can be a frame already returned — the
program that stashes (slice local 0 2) in a global slice and views an element
compiled and segfaulted with no diagnostic. The arm now recurses only when
the target's own type is an Array.
With it, the refusal/acceptance pair in test_flan.ml (one word apart) and a
view over an element of a global array in dyn-view.flan's mode 0.
The element check now runs before the lifetime check in all three container
arms: a local (Vec string) was told to make it a global, and a global
(Vec string) is refused anyway, so the advice was a dead end.
And the four strings that claimed more than the code does. flan_dyn.h
already had the honest version — a view is exactly as stale-safe as the
thing it is a view of — so the refusal message, box's comment and FIX.org
now say that instead of promising a dyn value can never dangle; a global
[i64] cut from a dead frame still passes and still reads it (ASan:
stack-use-after-scope in view_box). The element message no longer tells a
(Vec string) that string is not the case the restriction exists for.
dyn_ops.c's hand_vec comment no longer says flan_rt.c is unlinked when it
calls two of its functions; flan_rt.c said the same thing and is fixed too.
FIX.org's arena paragraph now separates the header's lifetime (compile time,
already covered) from releasing the arena under a live view: free-all traps
cleanly on the epoch, arena-destroy is a heap-use-after-free in
view_vec_check, the same gap flan_vec_check has on the typed side.
The dyn if truthiness review turned up that or's answer position, unlike
and's, still traps on a non-bool dyn value: or's short-circuit sentinel
sat in the then arm of its own if, the one check_if types first, so that
sentinel decided the whole expression's type and a later non-bool dyn
answer hit the strict bool boundary and unboxed itself into a trap
rather than surviving as itself. (or nil "x") — the canonical Clojure
(or x default) idiom — crashed instead of answering "x", identical on
all three backends.
or now binds its test to a temp and answers the temp itself, exactly the
way Clojure's own or macro expands: (or a b) becomes (let [t a] (if t t
b)), not (if a true b). The temp evaluates a once and lets the answer be
a without writing it a second time as the then arm; it is the temp's own
type check_if sees first, so or hands back the actual truthy operand the
same way and always has. Verified real output, unchanged, on LLVM, -O0
and --x86, and the survey program now exercises the case its own header
used to exclude for being unsafe: a non-bool value stopping or and being
handed back as-is.
check_truthy also gets three corrections a closer look found. Its own
[loc] used to come from the enclosing if/while/not rather than from the
condition itself, so the rt call and cast it builds carried the wrong
column in an --x86 disassembly or the dev inspector whenever the
condition was not the form's first token; it now takes loc from the
scrutinee's own AST node, confirmed against a real --x86 dump. A comment
now names the precondition its exception-swallowing retry rests on: none
of check.ml's save-restore sites (barrier, in_frames, in_defer, loops,
scope) are exception-safe, which is harmless only because the retry
always either succeeds cleanly or re-raises and aborts the compile
before ctx is read again — and would stop being harmless the day some
want-sensitive elaboration on this path could succeed differently on
retry. And a bare keyword condition, which used to be checked with
want:Bool from the start and refused by the keyword arm's enum-or-refuse
case, now resolves as the dyn keyword instead and is unconditionally
truthy — a deliberate loss of that diagnostic, the author's call, pinned
in test_flan.ml so it does not regress by accident.
The two typed-refusal messages captured before this pass (a float
literal condition, an i32 while condition) are unchanged, checked again
against the same baseline. test_flan.ml's parser test for or's shape is
updated to match the new let-bound desugaring.
A (Vec T), a slice or a fixed array crossing into dyn no longer refuses; it
is a view, one word in the box, over the container's own storage. Reads box
the element on the way out; writes tag-check the dyn value's tag against the
element type on the way in and trap, by name, on a mismatch, never coercing
or silently storing.
The open question the decision left — whether the descriptor points at the
container or snapshots pointer and length beside it — is settled by kind. A
Vec view holds the address of the Vec's own header (flan_rt.c's flan_vec,
restated in flan_dyn.c under the file's standing "if either table changes,
change both" rule) and reads ptr and len live on every operation, so a push
that reallocates cannot leave it stale: flan_vec_grow overwrites that same
header in place, and there is nothing captured at the crossing for the
growth to invalidate. A slice and a fixed array cannot grow, so a flat view
snapshots data and length once; pointing it at the value's own slot instead
would be worse, since a slot's lifetime is not the slice's.
The element set is i64, f64 and bool, not everything box already handles
typed-to-dyn. A string element's dyn form is a pointer into the collector's
heap, and a typed container's storage is arena or stack memory the collector
never scans — a wider set would let a write plant a live reference nothing
ever traces, which no care at the write site closes. (Vec string) and a
typed (Map K V) keep the "does not cross into dyn yet" refusal, now for that
reason.
flan_dyn.c gains a fourth object kind, OBJ_VIEW, and flan_dyn_len/at/set_at/
push and the printer each grow one branch for it beside the existing vec
one. A view's own stale-container check is the runtime's own spelling
(flan_trap, park-and-inspect) rather than flan_rt.c's rt_die, per the
duplicity doctrine; growing a Vec through a view calls flan_rt.c's own
flan_vec_push rather than re-implementing doubling and allocator adoption a
second time. (set (at target i) x) against a dyn target — a plain dyn vec or
a view alike — was a hole in the base dyn milestone rather than something
item 3 introduced; it is wired to flan_dyn_set_at here because a view's
writes needed it to exist at all.
Both backends: emit.ml and x86.ml both already passed a Vec or a Map to a
runtime call by address rather than by value; a fixed array crossing into a
view needed the same arm added in both, for the same reason — a copy would
view the copy and never see a write to the caller's own array.
test/dyn_ops.c drives the runtime directly with a hand-built Vec header and
a plain C array, ahead of any compiler involvement: reads, writes on both
element kinds, the tag-check refusal on every element kind, the range
refusal, and the push that grows and moves a hand-built header out from
under the view watching it. test_flan.ml turns the old "does not cross into
dyn yet" refusal into acceptances for Vec/slice/array, keeps it for a string
element and for Map, and adds the element-restriction refusal by name.
test/programs/dyn-view.flan is the compiler-level survey: a Vec view mutated
through both sides including the grow-and-move case, a fixed array's and a
slice's views, a bool Vec's view, and its own two trapping modes for the
acceptance rows to run against. test_sanitize.ml carries the survey's happy
path; test_dyn.ml's new refusals are the runtime's own.
A dyn scrutinee is no longer required to already be a bool: it is tested
for truthiness, Clojure's rule, not C's or Python's — nil and false are
the only falsey values, and everything else, including 0, "", an empty
vec, an empty map and a keyword, is truthy. A typed scrutinee is
unchanged and keeps needing a strict bool.
The runtime side is one new entry point, flan_dyn_truthy
(runtime/flan_dyn.c/.h), reading the tag directly rather than unboxing —
it never traps, unlike flan_dyn_need_bool. Both backends reach it the
same generic way flan_dyn_need_bool already did: check.ml emits an
ordinary Rt call plus the existing i32-to-bool Cast, so emit.ml only
needed the LLVM declare added and x86.ml needed nothing at all.
check.ml's check_truthy is the one funnel every boolean position in the
language goes through: if's own condition, while's, and not's argument.
when and cond reach it for free because they desugar to Ast.If in
parse.ml, and so does and's condition; or's condition does too, but its
answer position is a separate story — its short-circuit sentinel is the
then arm of its own if, which check_if types before anything else, so a
non-bool dyn value reaching that position still meets the strict bool
boundary. and's sentinel sits in the else arm instead, so the real
value's type wins and and hands back the actual last operand,
Clojure-style; or does not get that for the reason above, and reordering
it is a decision for another day, not this one. shortcircuit in parse.ml
carries the note.
check_truthy checks the scrutinee with no expectation first, so a dyn
value takes the truthy path and everything else takes the strict one. A
refusal on that second path is re-checked with the old want:Bool rather
than reported from the bare check, because a bare integer or float
literal, or a bare None, answers "what type is this" differently than
"is this a bool" — check.ml's own arms only give the nicer sentence
("expected bool, found the integer literal 5", "expected bool, found
None") when asked the second way, and that sentence is preserved exactly,
letter for letter, against what a typed if already said.
test/programs/dyn-if-truthy.flan surveys every falsey and truthy case —
nil, false, true, 0, a nonzero number, an empty and nonempty string, an
empty and nonempty vec, an empty and nonempty map, a keyword — through
if, when, cond, and, or, not and while, with real output pinned in
test_acceptance.ml across LLVM, -O0 and --x86. test_flan.ml covers the
checker side directly: a typed if still takes a bare bool and still
refuses a non-bool scrutinee and a bare None with their original
messages, a dyn if/not/while/when/cond/and/or all accept a non-bool dyn
condition. test/dyn_ops.c gets a matching set of direct calls to
flan_dyn_truthy, keeping the header's own contract with the C side.
(Some nil) at an already-(Option T) want reported expect's bare-T
sentence instead of its own: checking the argument against inner's
element type routed a literal nil through expect's "wrap the type in
Option" refusal before Some's own is_nil_lit guard ever ran, and the
advice was nonsense there — the type already is one. The argument is
now checked with no want when it is syntactically nil, which is what a
bare nil resolves against on its own, so it arrives at Some's own
check still dyn and still nil.
no_fallback_slots cannot see the slot unbox_option mints: %dx/%ax are
the pool-ran-dry fallback for a temporary root_plan counted, and a
named local root_plan never counted emits neither mark. The comment
where nil-option.flan and some-nil.flan were added to that list said
otherwise; corrected to say what the check does and does not cover,
and to record that the slot was verified by reading the IR directly
instead — flan.unbox-opt's dyn slot is pushed, flan.as-dyn's is a
plain alloca correctly, since its element is always scalar there.
dyn_offsets falls through Option/Vec/Map with no arm of its own,
correct today only because Check.hidden_dyn refuses a dyn inside any
of them at every storage site first. Commented at the fallthrough,
naming hidden_dyn as the gate and the typed-container view (M2 item 3,
in review now) as the kind of change that could relax it for Vec/Map
without anything here pointing back.
Both directions of the boundary go through expect, the way every other
dyn crossing does. A dyn's tag decides which case an (Option T) becomes
on the way in; an Option's own tag decides nil or a boxed payload on the
way out. box_option/unbox_option build the same If-over-a-tag shape get
and map-remove already build for the same reason, reading an Option's
tag and payload with the raw Field access Render's structural printer
already uses — nothing new for either backend to lower. A literal
Some/None skips the runtime check entirely, since the checker already
knows which case it is.
A bare T has no None to become. The literal nil the checker can see is
refused right there, at compile time, in expect itself — the author's
decision to do both halves rather than settle for the runtime trap
alone. Everything one step removed from the syntax — a dyn that only
turns out to be nil once the program runs — reaches flan_dyn_need_i64's
existing DynType trap, unchanged; there is no dataflow in this checker
for it to be otherwise (see "Ownership tracking repealed").
(Some nil) is refused the same way: the literal at compile time, with a
message saying why nil and None would collide; a dyn that turns out to
be nil only at run time through the new flan_dyn_need_not_nil, which
traps by the same route flan_dyn_need_i64 does.
(Option (Option T)) does not cross either direction — boxing Some of an
inner None would box it as nil, indistinguishable from the outer None,
the same ambiguity (Some nil) is refused for. The type itself stays
legal on the typed side; only the crossing does not exist for it.
(Option dyn) needs no case of its own in the boundary code — the
payload is already dyn, so box_option/unbox_option treat it as the
identity — but it is not yet a value a program can hold anywhere. The
per-type-descriptor pass (M2 item 2) refuses it at every storage site
today, the same way it refuses (Vec dyn), because a struct's dyn fields
are marked by byte offsets and (Option dyn)'s payload has none. Item 4
does not lift that gate; it only makes the boundary already correct for
the day items 2/3 do.
expect grew a ctx parameter to build the fresh slot the two new
crossings need — every call site threaded through, one context
mismatch caught and fixed in check_fn's tail-expression case along the
way. var's None case grew a direct Dyn arm: None at a dyn want is nil
outright, with nothing to build.
nil-option.flan carries the crossings that succeed and ends on the
bare-T trap; some-nil.flan is (Some nil)'s run-time half, kept in its
own file the way dyn-boundary.flan is one trap per program. Both are
in no_fallback_slots and test_sanitize.ml: the new dyn temporary
unbox_option's tag test mints is rooted, and reads its Option's tag and
payload through ASan clean, --sanitize matching the unsanitized run
byte for byte.
A worker's own pid was not enough to keep two variants of the same program
from colliding: raylib-image.flan and raylib-audio.flan export through the
raylib FFI to a fixed absolute path under /tmp, baked into the program
itself rather than passed in, so a default row and its -O0 row running as
two different workers could still race each other's file. Measured at a
~1.7% flake before. Both clusters now run through outputs_sync, off the
pool, the same as slurp.flan and files.flan already were for the same
reason.
The watchdog's own cleanup had two gaps. First, an inflight entry was popped
off the queue before the parent blocked reading its pipe, which is exactly
the moment a hang leaves the queue blind to the one worker it needs to see.
The entry now stays visible until its result is actually in hand. Second,
killing a worker's pid did not reach its clang, spawned three processes
away through Sys.command -> sh -> clang, which a forced alarm could leave
running as an orphan. Each worker now calls setsid on the way in, so it and
everything it goes on to spawn share one process group, and the cleanup
hook signals the negated pid to reach the whole group in one call. Eight
repeated forced alarms, zero processes left behind afterward.
Two comments claimed more than the code delivered. The order rows print in
is fully deterministic but is not serial — an inline FAIL still prints the
moment it happens, while a pooled one can sit behind up to eight other
submissions first — so the comment now says reproducible rather than
serial. And outputs_sync's isolation from the pool was true only because
nothing pooled happened to touch the same paths; a Pool.drain_all () now
runs before each of the four unpooled clusters, so that isolation holds
regardless of what gets added to the pool later.
And a worker now removes its own Build.workdir on the way out when its row
passed, rather than leaving one directory behind for every compile it ran
on top of what a single serial run already left. Left in place on failure,
where the original reason that directory is never swept — findable by name
when something is wrong with it — is exactly what the next person wants.
Reverified after all four: the raylib probe clean across twelve full runs,
the forced-alarm probe clean across eight, dune test read for FAIL lines
rather than trusted by exit code, three standalone runs byte-identical.
A descriptor's symbol was the type's printed form with every character an
assembler would refuse replaced by a dot, and the table was keyed by that.
The mangle is many-to-one — a Flan name may hold -, +, *, ? and / — so row-a
and row+a were one entry, the second of them was pushed with the first's
descriptor, and the collector read at another type's offsets: past the end of
the object when the first was the larger, and never where the second's dyn
actually sat. ASan named it, a stack-buffer-overflow inside gc_mark_all. It
is the same corruption root_plan pools its temporaries to avoid, arriving
through the name rather than through the supply, which is a lesson about where
identity lives: the table is keyed by Types.to_string now, which is an
identity, and the symbol carries a counter so two types cannot collide however
they mangle. dyn-struct.flan grows the pair, held live across the churn, and
an acceptance assertion asks the emitter directly how many descriptors it
wrote under that label — two, or the two are sharing one. That assertion is
the half with teeth: whether an overread off the end of a frame slot lands on
anything is luck, and the run's own output was not red under the defect.
The cap on a flattened array's offsets was bypassable by the thing it was
meant to stop. [4611686018427387904 S] wrapped the multiplication negative,
so the test read as under the cap, the declaration was accepted, and the
emitter then sat building the offset list until something killed it. A
refusal that overflows into an acceptance is worse than no refusal. The count
saturates at one past the cap now and the message says more-than rather than a
figure that came out of a wrap.
dyn_ops.c's second assertion had no teeth: a marker never writes through a
root, so "the word at a non-dyn offset is untouched" passed under any marker
at all. What discriminates offset-driven from word-driven is a dyn word the
descriptor leaves out, holding five hundred objects, that must NOT survive —
and it is checked by adding its offset to the table and watching the line go
red.
dyn_anywhere descended through Ptr and Slice, so (Vec (Ptr Cond)) was refused
with a sentence about a dyn inside a type whose storage contains none. A
vector of pointers to condition structs is an ordinary thing to write. It
stops at a pointer now, which is the line hidden_dyn already took for a bare
(Ptr S) and the line the whole argument rests on: a pointer is a view of
storage something else roots.
Which leaves the one honest hole, and it is named at the boundary where it
opens rather than left in a comment. Storage C hands back was never rooted
and never will be, so a (Ptr S) crossing a declare with a dyn anywhere under S
is refused by name — the same sentence a bare dyn already gets there, one
level down.
dune test --force: green, 0 failures. dyn-struct.flan clean under ASan and
UBSan and identical at -O2, -O0 and --x86.
The crux was never where to put a descriptor; it was how an instance finds
one. A bare struct on the stack has no header to hang a pointer off, and
giving it one would change the layout C interop agrees on, change the stride
of an array and change what embedding a struct in another costs. So it has
none. The instance never carries a pointer to its type and the collector
never derives one from the bytes: the pairing of an address with a descriptor
is made at the *push*, by the code that put the value there and therefore
knows its static type. That is the same trick the shadow stack has always
used, and it makes the stack case the easy one rather than the impossible one.
A descriptor is the size of an instance, a count, and a table of byte offsets,
emitted once per type as private static data. Flattened, not a graph — a
struct held by value contributes its offsets shifted by where it sits, and a
fixed array contributes its element's once per element — so nesting costs
nothing at run time and there is no recursion in the marker. The offsets of a
big array would be a big table, and that is capped with a sentence rather than
half of the repeat form item 3 will bring.
Four places a value of such a type can live, and all four are rooted: a frame
slot, a global, the temporary a call's by-value return is spilled into, and
the slot a condition that is not a place is evaluated into. The last two are
new and are the ones that were not obvious. A callee roots its dyn words and
pops them in its epilogue, so between the return and the caller's store the
only copy is a register, which a collector that finds its roots by address
cannot see; the same hole was open for a Flan call answering a bare dyn and is
closed here too. And a condition crosses as a pointer into the signalling
frame while a handler allocates, which is exactly what the original refusal
said could not be made safe.
dyn_roots grows into root_plan and both backends read it, which is what the
older note about one counter deciding both ends was always for. The aggregate
temporaries are pooled by type rather than handed out in mint order: a
positional supply that drifted would pair an address with another type's
descriptor, and marking arbitrary offsets off a base is corruption where a
missed root is only a bug. Pooled, the worst a drift can do is run out.
What is still refused is a dyn no static offset can reach — inside a typed
container, in a data type's payload or a union's members where the cases
overlay, or under an Option where the payload exists only beneath the tag.
A (Ptr S) and a [S] are deliberately not on that list: neither owns storage,
and the only storage this compiler hands out for such a type is a frame slot,
a global or a fixed array in one, all of them already rooted. That is what
lets a handler clause take its (Ptr Cond) and read a dyn payload.
test/programs/dyn-struct.flan is the evidence. It runs forty thousand rows
past flan_dyn.c's one-megabyte floor, so marks and sweeps really happen, and
it holds live values through them in all four places at once. It has teeth:
with the descriptor walk stubbed out of the marker, the kept vector's length
comes back 24 instead of 628 and its first element is a stale word. Clean
under ASan and UBSan, same output at -O2, -O0 and --x86. dyn_ops.c grows an
aggregate-root mode so the runtime half can be wrong on its own, with a
header word holding a bit pattern that looks boxed and is not a dyn slot.
--no-gc still refuses, and had to be told how: a struct with a dyn field is a
collected value even when no expression in the program ever has the type dyn,
because a zeroed one still has a word the collector is asked to mark.
dune test --force: green, 0 failures across every suite.
They were the whole of this binary's wall clock, and none of it was CPU the
machine did not have: one core busy, fifteen idle, for 30-odd seconds of
independent programs run strictly one after another. A small pool now forks
up to eight of them at once and reads results back through a pipe in the
order the rows are written in, so the log is byte-for-byte what it always
was — just produced faster.
Eight, not sixteen: dune already runs this suite's eleven binaries at once,
and test_dev's daemon is the actual critical path at roughly 46s against
this one's 30. Handing the pool the whole box would rob the binary that
matters more to speed up the one that matters less.
Every temporary path a worker touches now carries its own pid, so a
program's -O2 row and its -O0 row, run in different forked processes, can
never collide on the same exe or the same captured stdout underfoot. Three
rows could not be pooled that way regardless: slurp.flan and files.flan
write to fixed relative paths their own variants share, with a clean()
between them that only makes sense run in order, so those three stay on the
direct, unpooled path the whole file used before.
A worker that dies without ever writing its result — signalled, OOM-killed,
segfaulted mid-compile — is counted and named rather than silently read as
a pass; the exit status is trusted over a payload that never arrived. And a
worker clears its own SIGALRM handler on the way in, so the parent's
watchdog can never be triggered from inside a child holding a stale copy of
its sibling list. If the watchdog does fire from the parent, it now kills
every worker still in flight before it exits, so a hang does not also leave
orphans behind it.
Standalone, test_acceptance went from 43.4s to 13.8-20.8s across three runs
with identical output every time. The full dune test did not move — 49.4-
49.9s after against 49.8s before — which is the eleven-way sibling
parallelism already claiming what this pool would have used.
The report was that handler-case segfaults in a top-level global
initialiser. It does not, and never did. What crashes is one frame
further down, in a position that has nothing to do with startup:
(defn read-file [path string] dyn
(let [src (slurp path (heap-allocator))]
(defer (free src))
(read (as-slice src))))
slurp signals FileError, a handler further out unwinds, and the
transfer leaves this frame through its defers -- all of them. But src
was never written: the form that would have written it is the form
that transferred. free then reads whatever the stack held under that
slot, which at -O0 in a small program is zero and at -O2 is a live
pointer, which is why the same program looked like an optimiser bug
from one direction and a startup bug from the other.
return never had this. The checker splices the defers registered above
it and no others, and says so where it does it. The transfer exit took
the whole list, because it is one landing block per function and
nothing in the IR said where each defer had come into being.
So the count is kept. The first defer in a function mints an i64 slot
zeroed at the top of the body; each defer leaves a store of its own
number where it was written; and fdefers -- the transfer path's copy,
and only that copy -- tests the count before running each one. The
normal paths are untouched and still need no test. It is all in the
checker: what reaches a backend is a slot, a store and an if, so
neither emitter learned anything and the x86 one needed no frame of
its own.
test/programs/init-conditions.flan is the survey. The top half is the
part of the report that was never true: handler-case with its
condition firing and with its body completing, handler-bind, and a
restart-case, all four in a global initialiser, all four answering
what they answer anywhere. The bottom half is the part that was: a
defer below the signalling form, which must not run, beside one above
it, which must -- a fix that took the unregistered one off by taking
them all off would have traded the crash for a leak, and 302/2 is the
line that would catch it. Three acceptance rows, LLVM, -O0 and --x86.
The detector was valgrind, not ASan: reading a stack slot nobody wrote
is not ASan's bug class and it reported nothing on the broken binary,
while memcheck named the conditional jump in flan_vec_free with the
unwinding frame directly above it. The program is in both lists --
test_valgrind.ml because that is what saw it, test_sanitize.ml because
that is where the transfer exit's frames are already watched.
spec-conditions.md section 5 now says which defers a transfer runs.
The author's (defvar game-data dyn (handler-case (edn/read-file ...)
[(FileError [c] nil)])) works.
check_struct's comment claimed a nonempty map is never mistaken for a
struct literal; is_struct_map (parse.ml) says otherwise for any size of
.field-first map, not just {} -- (g {.x 1}) on a function g still reported
bare "unknown struct g" with none of the help {} gets. Extended the
helpful message to that shape too rather than only fixing the comment,
since it is the same trap one shape over. The two shapes need different
advice, not the same one reworded: an empty map can be bound to a variable
and passed as an ordinary argument, (let [m {}] (g m)); a .field-keyed one
cannot, because expr itself refuses a bare {.field v} outside a
struct-literal position, so there is no let-binding that rescues it. The
message for that shape says so instead of repeating advice that would not
work.
Watchdog.is_dying is gone. The reviewer's own probe settled it: forcing
watchdog.ml's dying to flush_all and Unix._exit 2 directly, bypassing
at_exit entirely, gives the same exit 2 with no shared flag, no LIFO
assumption, and no window for a raise between setting a flag and exiting
to leave failures recorded and the process at exit 0. test_acceptance.exe
has no other at_exit registration (grep confirms), so nothing depends on
the watchdog path running through that chain.
test_acceptance.ml's own comment claimed calling exit inside an at_exit
handler recurses through do_at_exit. Checked directly against this
compiler (OCaml 5.2.0): it does not -- each handler gets a run-once guard
since 4.14, two stacked handlers with the inner one calling exit 7 both
ran exactly once and exited 7. Unix._exit is kept, but for the true
reason: it is not about correctness, it is about being the last word --
_exit terminates immediately and skips whatever the rest of the at_exit
chain would otherwise still do, so a handler this file grows later cannot
change the outcome underneath this one.
test_flan.ml's loose needle "keyword/value pairs" matched both the
odd-number-of-forms message and the wrong-key message; tightened to
"keyword/value pairs — found", which only the row's actual message
contains.
And {:where 1} as a defn's whole single-form dyn-map body now refuses,
where {:a 1} and {} in the same position do not -- :where is peeled
unconditionally, with no single-form exception, because that is what lets
it catch a moved closing paren leaving a stray predicate as ordinary body
code. Deliberate, and now pinned, so the next edit to this arm has to
notice it is choosing to narrow the language again rather than finding out
from a bug report.
Verified every case by compiling, including the two-handler exit(7) probe
run directly against this OCaml. dune test --force: exit 0, clean grep for
FAIL and Fatal error.
F3: the unknown-struct-vs-function message overgeneralized a one-case
parser quirk into a language rule that does not exist. (g {:a 1}) compiles
fine -- only the empty map is stolen by is_struct_map's Map [] -> true.
The message now names {} specifically and says why: it is read as the
zero-field struct literal, not "a map literal cannot be passed as an
argument".
F2: the ordering refusal named machine numbers only, when Types.is_comparable
also admits enums and enum-compare.flan orders one with (< k :mid). Both
messages -- equality and ordering -- now name every type each actually
covers.
F1 and F5, together, since both live in the same parse.ml arm: the
rest = [] carve-out that let a single-form dyn map body through
reintroduced the exact bug the earlier fix was for. (defn mx [a $t b $t]
$t {:where (ordered? $t)}) -- a :where clause with nothing after it, what
a moved closing paren produces -- no longer matched, fell through to expr,
and printed "unknown function ordered?" from inside what was meant as a
predicate. A :where map is peeled unconditionally again, whether or not
anything follows it; every other keyword, plus the empty map and any
non-keyword key, is still peeled only when the body has more after it, so
a real single-form dyn map body is still left alone. The comment
justifying the discarded-map refusal claimed a map literal has no side
effects; it does -- {:a (println "hi")} prints, confirmed by running it --
so the reasoning now names the actual problem, a value going unused, not
a false claim about purity.
Closing that F1 hole this way opened two more, both traced by rebuilding
the pre-fix parse.ml and diffing real compiler output rather than
reasoning about it: {.x 1} at a defn body's head used to get its own
message, "a bare map is not an expression", and briefly started getting
"a constraint map is keyword/value pairs" instead, because the broadened
guard no longer required a keyword and started catching the struct-field
shape too. Excluded explicitly, with a comment saying why, so expr's
dedicated diagnostic fires again. And (defn main [] i32 {} 0) and
(defn main [] i32 {"a" 1} 0) -- the empty map and the non-keyword-keyed
map, the two siblings a keyword-only guard could never have caught --
now get their own refusals alongside the keyword case, each pinned with
a rejects_check row, along with the where-with-no-body case and the two
legitimate single-form bodies that must keep compiling.
F6: the acceptance runner's tail already caught every failure on every
path through the binary -- there is no skip branch that bypasses it, and
the no-clang branch runs zero rows -- so last round's at_exit guard and
the FIX.org note both overstated what was broken. Both now say what was
actually true: the exit status was already trustworthy, the guard is
insurance against a future case leaving past the tail instead of through
it, and the other nine test binaries were already sound the same way.
The guard also had a real bug of its own: forcing exit 1 whenever
failures was nonzero would stomp the watchdog's own exit 2 if a hang
followed a few already-failed rows, since Stdlib.exit runs at_exit
handlers LIFO. watchdog.ml now flags when it is the one unwinding, and
test_acceptance.ml's guard defers to it -- shared state for a single
caller, justified by there being no other way for one at_exit handler to
know a sibling handler is already mid-exit with a code of its own to
protect.
Verified every case in this commit by compiling and, where it mattered,
running the actual program -- not by inspecting the arm and assuming.
dune test --force: exit 0, clean grep for FAIL and Fatal error.
Constraints parsing peeled a body-leading map only when its first key was
literally :where; any other keyword fell through to the body, so a typo'd
key surfaced as a baffling error from inside what was meant as a predicate
and a stray map at body start compiled away silently. Any keyword-first map
is read as a constraint map now, but only when something follows it in the
body — a single-form map body is a real dyn value and not a discarded
statement, so that case is left alone.
An empty map literal still parses as a struct literal, (P {}) still meaning
the zero struct for a real struct name — the parser has no symbol table to
tell (take {}) apart from it at that point. check.ml now catches the case
where the name turns out to be a known function instead and says so, rather
than "unknown struct take".
flan_dyn.c's tag comment still said 6 and 7 were free; keywords and maps
took 4 and a kind field under BOX_OBJ, not new top-level tags, so 5, 6 and 7
are what is actually open for the interop handle. NEXT.md and json.flan both
still pointed at test/programs/arena-edn.flan, gone since edn/read stopped
taking an allocator; both now point at what replaced it.
flan_rt.c's flan_str_eq comment claimed the empty string literal was a
hypothetical null-pointer string; it isn't, its address is an interned
symbol's. The real case the zero-length guard exists for is a zero-length
container converted to a string. check.ml's ordering refusal said a string
has no comparison at all, which stopped being true when typed = and !=
grew strings in daed039 — split the message so an equality refusal and an
ordering refusal say the right noun, and updated the pinned rejects_check
rows to match. string-eq.flan gained the row the fast path most wants
tested, a slice against the prefix it was cut from sharing a base pointer at
different lengths, plus a != row at equal length with differing bytes;
acceptance now carries the real output, captured by running the program on
all three lanes. x86.ml's xor-1 comment now names the 0/1 return contract as
a requirement flan_str_eq must hold, not an incidental fact. SPIKE-DUPLICITY
now says plainly that its equality-and-ordering argument landed in daed039
and marks its transcript as the historical state that argument was made
against. FIX.org ticks M2 queue item 5.
And the acceptance runner: the tail check that turns a nonzero failure count
into exit 1 was already there and already fired — a fresh build with one row
broken already exited 1 before anything here changed. What wasn't proven is
that every path through the file's clang/wasmtime/raylib/lldb probes still
reaches that tail rather than skipping past rows that already failed. An
at_exit guard now closes that class regardless of which path the process
leaves by, flushing stdout first so a failing run's FAIL lines survive
Unix._exit rather than being dropped from the buffer. Verified both
directions with a deliberately broken row: dune test exits nonzero and the
log still carries the FAIL line and the failure count; restored, the same
run is exit 0 with nothing printed but green summaries. The other test
binaries were checked for the same gap and none have it — each gates its
own exit on a single failures ref that the tail already reads.
Every mark it left was an addition, and addition commutes, so a backend that
ran the defers outermost-first produced byte-identical output and the row that
was supposed to be watching the order could not have told. The claim was in the
comments and not in the numbers. cleanup.flan already had the device for this —
a shift rather than a sum — so the log here is a digit trace now, and the two
frames under a catch read 12 where a wrong order reads 21.
Rewriting the trace made room for the three behaviours that worked and nothing
pinned. A return inside a clause is an ordinary return from the function that
wrote the form, because that is where a clause runs: it leaves through the
function's own exit, runs the defer registered there after the two the unwind
already ran, and leaves the handler stack empty behind it, which the bare
signal that follows in main is the check on. A defer inside a clause is refused
for the reason every nested form is refused one. And a handler-case inside a
defer works, because a defer may not start a transfer that leaves it and this
one begins and ends its own.
The program is registered with the sanitizers, where the interesting failure is
not the heap but a handler or restart frame left on a stack pointing into an
alloca that has gone — an output comparison cannot see that until something
much later calls through it. It is clean; it was also leaking sixteen bytes out
of the vector main allocates to prove the allocator context came back, which is
the test's own litter and is freed now.
docs/PORTING.md ranked handler-case as one site handler-bind covers. It still
is one site, and handler-bind still covers it, but it is no longer the closer
translation: a catch block is assumed everywhere it is written to see the
locals around it, and only the clause that runs at the form does.
The handler clauses are lifted left to right rather than by List.map, whose
order is unspecified. Each lift names itself after the count already on the
list, so an order nobody chose would number the clauses of one handler-bind
differently between builds, and those names go into a redefinition module.
The unwinding handler, which spec-conditions.md named and left unwritten while
it asked whether the thing should be a macro over the two operators that were
already here. It should. (handler-case B [(T [c] A)]) is checked as
(restart-case (handler-bind [(T [c] (invoke-restart 'R c))] B) (R [c T] A))
with R a name the form makes up for itself, which is Common Lisp's own
definition of the operator and means neither backend needed a line.
What that buys is not economy, it is the correctness of the parts nobody can
see. The defers between the signal and the form run, and the allocator a
with-allocator rebound is put back, because a transfer already does both for
every frame it leaves. The body and every clause agree on one type, because a
restart-case's body and clauses already do, and a clause that disagrees is
refused with the same message an if with disagreeing arms gets. A condition no
clause lists installs no frame that matches it and goes on outward untouched.
A clause sees the establishing function's locals, which a handler-bind clause
cannot, because a restart clause runs where it was written.
The body comes first and the clauses after it, the opposite of handler-bind's
order: one reads as something put around a body and the other as a body with
answers hung off the end of it. The restart the two halves meet over is named
after the function and numbered within it, and it has to be unique per form,
because two nested handler-cases sharing a name would have the inner frame
shadow the outer one and land a condition at the wrong place.
The refusals name handler-case rather than the machinery underneath it, which
is why check_handler_bind and the restart clauses now take the word the reader
wrote. A break loop entered under a handler-case still lists the made-up
restart, and taking it there is refused loudly rather than answered wrongly;
hiding it would mean a field in a frame layout spelled out in three files.
The survey program runs the same under LLVM, at -O0 and under --x86: normal
completion, a caught condition, one nobody listed passing through with the body
carrying on, both defers on the way out, the two nestings against handler-bind,
a clause that signals and is caught outside the form it belongs to, and a
with-allocator whose restore is on the transfer path.
Types.is_equatable splits from is_comparable: a string answers equal?
now, bytewise, but still answers no to ordered? — there is no collation
the language has picked, so < and friends keep the refusal they had.
The comparison itself is one new runtime entry point, flan_str_eq
(runtime/flan_rt.c), length-mismatch and same-pointer fast paths ahead
of the memcmp, called identically from both backends: emit.ml pulls a
string's ptr and length out of the %slice SSA value and calls it
directly in the Eq/Ne arm; x86.ml adds an arm ahead of the generic
scalar comparison that reaches it through call_native, flipping the
answer for != the same way Not already flips a bool.
test_flan.ml covers the checker side directly and through a generic
instantiated at string, including the two different ways ordered? and
equal? fail at that type. test/programs/string-eq.flan is the survey
program — same pointer, differing lengths, equal content at distinct
addresses (a literal against a fresh heap string), a difference in the
last byte, and the empty-string cases — with acceptance rows for LLVM,
-O0 and --x86 in test_acceptance.ml.
The dyn runtime gets a map object and an interned keyword, alongside the
vec it already had. {:a 1 :b s} is a map literal wherever a struct
literal isn't — the parser tells the two apart by whether the first form
in the braces is a .field symbol — and a bracket literal builds the
runtime's own vec rather than a typed array wherever a dyn is wanted, which
is what lets a map literal's values nest arrays and maps freely. get, put,
len and has-key? all learn a dyn-map arm alongside the typed-map one they
already had, and (keyword s) builds the same interned value a :foo literal
does, for a name that only exists at run time. nil is now a literal, the
dyn absence value that get answers for a key a map does not hold.
On the runtime side, flan_dyn.c gets an OBJ_MAP that shares the vec's
storage arm and doubles its accounting, a linear-scan intern table for
keywords that makes equality an identity compare, and structural map
equality by lookup rather than position. The marker traces a map's
interleaved keys and values the same way it already traced a vec.
edn/read and its callers move off the old (Option Value) union entirely:
a document is plain dyn now, sets are dyn maps to true, and arena-edn.flan
is retired along with the union it demonstrated. The acceptance suite's
edn-read and json rows were recaptured against the new shape, and a new
dyn-map.flan program exercises the map and keyword operations end to end,
including a 200k-iteration churn loop against a rooted map that runs
GC for real, across the LLVM, -O0 and x86 rows, and under the sanitizer.
Keywords are dyn everywhere an enum isn't expected, which changed what a
couple of existing checker tests actually see refused; both were updated
to the sentence the checker gives now rather than the one it used to.
Found by adding the --x86 rows beside them: the new rows failed, and so did
the LLVM rows they were copied from, identically and on the tip. That is the
tell -- a backend cannot change what a runtime prints, so the expectation
was what had gone stale. Both were written while flan_dyn.c was the stub
that mallocs and never frees, and the real renderer landed with different
answers to two questions the stub never had to answer.
The container line has a space after the open bracket because the space is
a prefix per element rather than a separator between them, and a text nested
in a container is escaped and quoted while the same text printed alone is
not -- three bare on its own line, "three" inside the vector. Both are
deliberate and both are pinned by the runtime's own C test, which asserts
"[ 1 2 3]" and "[ \"x\" \"a b\" ...]"; this file is the side that had not
caught up, so this file moves.
The trap sentence is the same story: it names the tag it found and the tag
it wanted, and the row now matches on that half rather than on the wording
the stub used.
dune test --force: 6 failures to 0.
x86 is what flan dev takes by default and dyn is the iteration feature, so
a backend that refused dyn meant the two halves of the dev loop could not
be in the same program. The refusal was one arm of is_agg, and it said the
true thing: it was never the representation that was missing. A dyn is
uint64_t, a scalar in both calling conventions, classified by every rule
this file already had; every operation on one is a Tast.Rt primitive and
call_rt has always known how to make one of those. What the lane actually
cost was the collector's root discipline.
Which is emit.ml's, reused rather than rewritten: Emit.dyn_roots counts the
roots for both backends now, so the pushes and the pops balance because one
counter decides both ends, and the two backends root the same nodes because
there is one counter and not two. A zeroed frame slot per dyn slot and per
dyn-producing call, minted beside the channel and outside every scoped --
the bump allocator reclaims at the end of a statement and a slot minted in
the body would be handed out again while the collector still held its
address. Pushed from the body buffer, not the prologue's, because a call
clobbers the registers the prologue is still spilling from. And one pop in
the epilogue, which is the whole of why this backend needed no landing-pad
work for it: there is exactly one epilogue, and the return, the fall-through
and the transfer exit all arrive at it. emit.ml needs the same pop at five
separate rets.
The ABI point the dyn handoff left open for the integrator is settled by
reading the other side rather than by agreeing: flan_dyn.c's mark follows a
value only when the quiet-NaN prefix is set, and the zero word does not have
it, so a zeroed root decodes as the double 0.0 and is never an address
anything dereferences. Zero is safe for a reason. The header says so now.
And one line in dev.ml that was never x86's: the merged dev host resets the
condition stacks and the frame chain between runs, because main is
re-entered by longjmp and pops no frame -- and it never reset the root
stack, so every root a finished run pushed still named stack the next run
was about to write over. That gap was an LLVM dev build's too.
Verification, and one of the numbers is new. @x86: MATCH 129 -> 135, DIFFER
0, REFUSED 0 -- the five dyn programs off survey.sh's llvmonly list, which
is gone rather than empty, plus p13. dune test --force green, with --x86
acceptance rows beside the LLVM ones for all five dyn programs, dyn-boundary
asserted on the same exit 134 and the same sentence on both.
p13-dyn-collect.flan is the one that is not a formality. Nothing else in
this repository allocates past flan_dyn.c's one-megabyte floor, so nothing
else collects even once, so a program whose roots are entirely wrong passes
every output test there is -- the handoff wrote that about the stub and it
outlived the stub. p13 allocates several megabytes of garbage while holding
live values across it: at forty times the corpus size it peaks at 4MB of
RSS, which is the collector running many times over, and both backends
still print the same four lines.
The louder failure had the quieter answer. A generated reader accumulates
errors on the cursor — which is what lets it be a straight line of
assignments — and the cursor is made and dropped inside the entry point, so a
stray brace in a file read at run time handed the program a zeroed struct and
said nothing at all. That is the one thing the rest of vendor:edn refuses to
do: read-file answers an Option precisely so a malformed document is
distinguishable from one that is literally nil, and the hand-written reader in
test/programs/edn.flan tests ok? and prints the reason. ReadFailed is the
derived reader being as honest, in both packages, and it sits beside
SchemaDrift because both are "the file is not what this program was built for".
Then the writing-down. docs/BUILT.md gets the section: the four things the
macro system did not have and now does, each general and none of them
mentioning EDN — a macro reading a file at the call site's path, a package
macro calling its package, one call answering several declarations, and
compile-error, which is the one piece that had to go in the compiler and the
reason it had to. The set rule the real game file decided is there too, and
what defjson shares, which is the design and not the code.
NEXT.md item 9 and PORTING.md §3.9 both close. Not as (read-edn T bytes): the
struct comes from the *file* rather than from a type declared by hand, so the
~80 lines PORTING prices for two schemas are not written at all. The competing
answer PORTING names — compile-time embedding — turned out to be the other half
rather than a competitor: the shape comes from the file at compile time and the
bytes may come from an embed beside it, which is exactly what
test/programs/edn-provide.flan does.
vendor:json depends on vendor:edn for nothing, and borrowing a shape walk
across that line would be a dependency for the sake of a resemblance. What
carries over is the shape of the answer — one walk giving a type, the
declarations that type needs and the expression that reads one; a refusal
carried in a field rather than raised; a typed one-line constructor per
collection; and a compile-error wrapped in a defn nothing calls.
What is genuinely different is four things. Strings go through string-of and
never through .text: .text is the raw interior with escapes undecoded, so a
field read off it would hold a backslash and an n where the file meant a
newline — which is the first two lines of the acceptance output and the reason
they are two. Commas and colons are tokens rather than whitespace. An object's
keys are strings, so a key has to be refused when it is not a name a program
could write, and refused again when it carries an escape: the generated reader
compares against the bytes as written, which costs no allocation per key and is
only the same question when the name is written plainly. And there are no sets,
so there is no map-key path and no fixed array — every collection is a (Vec T)
and defjson is the smaller of the two by half.
JSON has no integer type; the tokenizer draws the line at whether a number has
a fraction or an exponent, which is the only line there is, so 1 derives i64
and 1.0 derives f64. That is the file's own distinction and the honest one to
take.
@x86 matches on it and @sanitize is clean.
A defer is in the typed IR twice -- spliced into the body for the normal path,
and again in fdefers for the path a transfer leaves through -- so a dyn
temporary inside one is emitted twice. dyn_roots counted only the body's, and
the second copy went into slots nothing had rooted.
Nothing failed, and that is the whole reason this is worth a commit of its own.
dyn_tmp falls back to a plain slot rather than unbalancing the stack, so the
pushes and the pops still matched, the program ran and printed the right answer,
and the values were simply invisible. Against a stub that never collects there is
no symptom to find -- no leak, no crash, no wrong number. It would have become a
symptom the week the real collector landed, in a defer reached only on a handled
condition, which is close to the worst place to start looking.
What found it was the IR: a rooted slot is spelled %dr and the fallback %dx, and
the assertion is that no dyn program in the corpus emits one of the latter. That
is now a test over all five dyn programs, and it is the only check in the lane
that can see a missing root while there is still nothing to lose one by. When
the collector arrives it is the thing to extend rather than replace.
Also checked, both clean: flan dev --llvm builds and runs a dyn program, which
is the route the x86 refusal sends people to and would have been a link error in
the worst possible place; and the daemon's own refusal already names the flag.
The promise is that this program carries no collector, and the way to keep it is
to refuse every dyn rather than to emit a different program: a dyn value is one
the runtime allocates and the collector owns, and there is no smaller version to
fall back to. So it runs between checking and emission, answers unit or raises,
and hands the very same program on. Emit has no field to branch on and is told
nothing.
That is what makes the byte-identity claim true rather than approximate, and it
is tested by compiling three annotated programs twice and comparing the text. A
field, a mode, or a comment that mentioned the flag would break it on something
incidental, a long way from anything to do with dyn.
Every site is named, the way the global cycle refusal names the whole ring: a
reader who has to annotate their program wants the list, not the first one and
then another compile. Globals and signatures as well as body values -- the two
files it is tested against report nine sites each, and the floors are set under
that so an added line does not fail the test and a pass that named one site and
stopped would.
The four programs run at -O2 and -O0. dyn-boundary is asserted on its exit
status as well as its output, because the boundary is only interesting in that
it can fail and a test that showed it working would be testing the easy half.
The x86 survey skips them by name: a REFUSED there means a node that backend has
stopped lowering, which is a regression, and this is the opposite -- a lane that
has not started. Take a name off llvmonly when the lowering arrives and the
survey will say whether it works. 128 match, 0 differ, 0 refused.
Checked while writing these: a dyn function with an early return pops its roots
on both paths, and one with a defer pops on the transfer path too.
edn-provide.flan reads assets/edn/tileset.edn through a struct derived from it,
and its first five lines are edn-read.flan's first five character for character.
Two readers over one file agreeing is what says the derived one is right; either
alone could be self-consistently wrong. The pair memberships are the derivation
deciding in public: the set became a (Map [2 i64] bool), so [3 4] is a key and
[9 9] is not, where a version that made it 108 loose integers would have
compiled and answered differently on all four.
A tuning file beside it covers the rest of the matrix — a string, an integer, a
float, a boolean, a vector summed rather than counted, and a map inside a map
read two field loads deep — and then drift: the struct was derived from a file
with :speed and without :level, and the bytes read carry the opposite. Both are
named, and the read carries on.
The refusals write their own data file, because the data file is the test. Each
is asserted on the position it names, not on the fact of failing, and one of
them checks a line and column into a file the compiler is not reading — which
is the whole of what compile-error was added for.
Two things the tests caught. Load extends the ambient macro set rather than
replacing it, so a package reached twice handed its declarations over twice and
the module refused them as a redefinition; Macro.compile dedupes by name, which
is the rule macro_union already applies a level up. And `where` held a line and
a column at once, which the prelude's note over append-i64 says cannot be done:
i64->bytes renders into one shared static buffer, and both numbers read as the
second one.
The !-means-mutates convention distinguished nothing — there is no
immutable counterpart to contrast with — so every mutating name drops
the mark: sort, sort-by, sort-bytes, swap, reverse, append, append-i64,
append-f64, encode-rune, split-next, map-remove, map-next, and the test
helpers beside them. Two could not simply shed it: map! is map-in-place,
because map is the into transform's word and means the non-mutating
thing; put! is put-at, because put is the Map builtin. The ?-means-asks
convention stays. Dated records keep the old spellings; watch.clj's
reset-spies! and the other Clojure names are not ours to rename.
The x86 backend ran initialisers from .init_array and the LLVM one refused
them by name, so (defvar frame Allocator (arena-new 262144)) — which the
author kept writing — was a program on one backend and an error on the other.
A rule that holds on one backend and not the other is not a rule.
The checker lifts a computed initialiser into a function of its own and the
global's initialiser becomes the call. That is what gives it a frame, which is
the bug underneath the feature: a `let` or a `match` in an initialiser indexed
a slot array of length zero and took the x86 emitter down with an uncaught
Invalid_argument.
Both backends call the lifted initialisers from main, after flan_rt_init and
before a line of the program's own code — Odin's __$startup_runtime shape, not
a constructor, so the runtime is up and the order is the compiler's to choose.
x86 keeps .init_array for one thing only, and it is named: writing the
constant image this backend has no folder for, which is standing in for the
other backend's object image rather than for a program.
The computed globals are sorted by what they read, transitively through the
functions they call, so a global written above the one it reads works and a
ring is refused with every name in it. A reload still re-runs nothing: a new
global with a computed initialiser starts as ZII on both backends.
The refusal that lived in x86.ml is now the checker's and is narrower. Nothing
can escape an initialiser — the handler and restart stacks are empty and every
frame it pushes it also pops — so what is refused is a signal or an
invoke-restart with no handler-bind or restart-case around it, which is inert
by construction. A restart-case inside one is ordinary code, which is what
makes (defvar data (Vec u8) (slurp "level.edn")) an ordinary program.
Three refusals go with the premise they rested on: a container global with a
computed initialiser, a union member in a defvar, and a data type case in one.
A defconst is untouched and keeps all three.
One change here is not about any of that. sand.flan carried an unfinished
line — (defvar game-data (embed (with-allocator frame ))), which parses as a
declaration whose type is (embed ...) — so the checker refused the file and
`dune test` was red at the tip of dev-loop before a line of this landed,
verified by stashing this work and rebuilding. It is commented out rather than
guessed at: the arena above it is the half that works, and what the global
should read is the author's to decide.
Two things the motivating line wanted and could not have.
or-else and some? are the first prelude family over (Option $t), and the
first that declares no {:where} at all: they move the payload out or read
the tag, and neither is an operation a type variable has to be admitted
to. So they instantiate at every type, including the ones that own
storage — where the answer is a header onto one of the two buffers and
the branch not taken is still the caller's to free, which the comment
says because "or a default" reads like it consumes the default.
none? is declined as (not (some? o)), and an unwrap that signals on None
is declined for the reason file-size is an Option at all: absence is a
reply and not a fault, and whether an empty one is an error is the
caller's question.
edn/read-file is worth having for one fact the package already argued:
every string in a Value is a copy, so the source buffer is dead the
moment read returns and nothing outside the call can be holding it. It
slurps against the heap by name — the one allocator this package names,
because the buffer's life is inside the call and is not the caller's
tier to choose — defers the free for the transfer path, and passes
slurp's FileError straight through with both restarts armed. Folding a
missing file into None would collapse the very distinction the Option
exists for.
There is no json/read-file and json.flan now says why: a Token's text is
a slice into the caller's buffer, so the prerequisite is a json/read
answering a self-contained document, and there is no Value type there to
answer with.
The defvar initialiser in the motivating line is still refused as
computed, so edn-read.flan writes it as a defn and says so; everything
inside the with-allocator is verbatim.
