flan/test/test_acceptance.ml
Joseph Ferano db7be70f7d Rounding from the one mode the language has, and sqrt from libm
floor, ceil and round over f32, which is what a position and a tile coordinate
are here. The only rounding mode available is the cast's truncation toward
zero, so each of these is that cast plus the correction the mode does not
make, and the content is which inputs make the cast itself undefined. NaN
fails every comparison, so it needs its own (not (= x x)) and nothing else
finds it; the infinities fall out of the magnitude test; and above 2^23 an f32
has no fractional bits left, which makes returning the input there the exact
answer and also the guard that keeps the cast inside i32.

round is half away from zero, written as floor of the magnitude and mirrored.
The obvious (floor-f32 (+ x 0.5)) is wrong twice: half-up rather than
half-away, so -2.5 comes out -2, and at the largest f32 below 0.5 the addition
alone rounds to 1.0 and answers 1 for a number under a half. Both are in the
table, which is why every case there is a negative or a half.

sqrt is the decision in this commit and it goes out to libm, which is a change
to the release link and so is said out loud. Every other number in the prelude
is reachable from the four operations and a cast; a square root is not.
Newton's method needs a starting guess, the good guess comes from
reinterpreting the exponent bits, and the language has only value-preserving
casts - no bit-cast between f32 and u32. Without one the iteration needs a
scaling loop to normalise and still produces a result that is merely close,
which is the one thing a standard library must not hand back. IEEE-754 makes
sqrt correctly rounded, so libm's answer is the same bit pattern on native and
on wasm32; for this function the byte-identical argument points at C rather
than away from it.

The cost is -lm on every link, and its placement matters. It goes after the
objects, not in the leading flags, because --as-needed drops a library named
before the object that wants it. Worse, at -O2 LLVM folds most sqrtf calls
into the hardware instruction and the symbol never has to resolve - so this
looked linked before the flag existed and failed only at -O0, which is exactly
why the table runs both. Untested against --target=wasm32: wasi-libc ships
libm.a as a stub because the symbols live in libc, so it should be inert
there, but nothing here exercises it.

The better fix is not in this lane. llvm.sqrt.f32 as a builtin in check.ml and
emit.ml is one instruction, no symbol and no flag, and it belongs to whoever
owns the compiler.
2026-09-11 19:43:07 +07:00

379 lines
18 KiB
OCaml

(* The milestone-2 acceptance test: a table of expression/result pairs run
through a compiled calc-me (plan.org, Build sequence).
It is a table rather than a golden file because milestone 3 runs the *same*
table on wasm32 — headless is what makes one test cover both targets. *)
open Flan
let failures = ref 0
let scratch = Filename.get_temp_dir_name ()
let run exe arg =
let out = Filename.concat scratch "flan-acceptance.out" in
let cmd =
Printf.sprintf "%s %s > %s 2>&1"
(Filename.quote exe)
(match arg with None -> "" | Some a -> Filename.quote a)
(Filename.quote out)
in
let code = Sys.command cmd in
let text = In_channel.with_open_bin out In_channel.input_all in
Sys.remove out;
(code, text)
let compile ?(opt = "-O2") ?(checks = true) ?(dev = false) path =
let exe =
Filename.concat scratch
("flan-t-" ^ Filename.remove_extension (Filename.basename path))
in
(* Through [Load], so a program with an (import ...) is buildable here: it
brings back the package's C shim and linker arguments as well. *)
let l = Load.program ~file:path (Parse.program (Reader.read_file path)) in
let p = Check.program l.Load.decls in
ignore (Build.executable ~opts:{ Build.default with opt; checks; dev }
~csrcs:l.Load.csrcs ~lflags:l.Load.lflags p ~out:exe);
exe
(* No Str, and the reader is hand-written for the same reason. *)
let contains hay needle =
let n = String.length needle and h = String.length hay in
let rec go i = i + n <= h && (String.sub hay i n = needle || go (i + 1)) in
go 0
let () =
match Sys.command "command -v clang > /dev/null 2>&1" with
| 0 ->
let exe = compile "../calc-me.flan" in
let case name arg expected_out expected_code =
let code, text = run exe arg in
if text <> expected_out || code <> expected_code then begin
incr failures;
Printf.printf
"FAIL %s\n got: %S (exit %d)\n wanted: %S (exit %d)\n"
name text code expected_out expected_code
end
in
let evaluates src expected = case src (Some src) (expected ^ "\n") 0 in
let rejects src = case src (Some src) "calc-me: cannot parse\n" 1 in
(* Arithmetic and precedence *)
evaluates "1 + 2 * (3 - 0.5) / 2" "3.5";
evaluates "1+2*3" "7";
evaluates "2*3+4" "10";
evaluates "(1+2)*3" "9";
evaluates "10/4" "2.5";
evaluates "7" "7";
evaluates " 7 " "7";
evaluates "1.5+2.25" "3.75";
(* Left-associative: 1-2-3 is (1-2)-3, not 1-(2-3) *)
evaluates "1-2-3" "-4";
evaluates "8/4/2" "1";
(* Unary minus, including nested *)
evaluates "-5" "-5";
evaluates "-(1+2)" "-3";
evaluates "3 * -2" "-6";
(* Whole input or nothing: trailing junk is an error, not ignored *)
rejects "1 +";
rejects "(1+2";
rejects "1 2";
rejects "";
rejects "+";
rejects "1+2)";
case "no argument" None "usage: calc-me \"1 + 2 * 3\"\n" 1;
(try Sys.remove exe with Sys_error _ -> ());
(* Programs whose whole output is fixed. These cover the milestone-2
surface calc-me does not reach — globals, 2-D arrays, places through a
pointer, casts, match with either arm taken, and the value semantics of
spec-memory.md. *)
let outputs ?opt ?dev name path expected =
let exe = compile ?opt ?dev path in
let code, text = run exe None in
if text <> expected || code <> 0 then begin
incr failures;
Printf.printf
"FAIL %s\n got: %S (exit %d)\n wanted: %S (exit 0)\n"
name text code expected
end;
(try Sys.remove exe with Sys_error _ -> ())
in
let values_out = "1\n5\nel\n" in
let machine_out = "12\n30\n2\n2\n3\n3.5\n42\n99\n12\n123\n" in
outputs "value semantics" "programs/values.flan" values_out;
outputs "machine surface" "programs/machine.flan" machine_out;
outputs "unit main exits 0" "programs/unit-main.flan" "ok\n";
(* The prelude's slice algorithms. Every assertion here is over an input a
wrong implementation fails: unsorted with duplicates, negatives and an
odd length; a reverse-sorted slice; and a sort of a subslice whose
neighbours must be untouched, which is the in-place, ptr+len claim
itself. At -O0 as well — a slice parameter is an alloca of a two-word
struct, and mem2reg is exactly what would hide it being copied. *)
let slices_out =
"5 -3 5 0 12 -3 7\n23\n-3\n12\n0\n-1\n99\n\
7 -3 12 0 5 -3 5\n-3 7 12 0 5 -3 5\n\
-3 -3 0 5 5 7 12\n1 2 3 4 5\n\
100 -1 0 4 9 9 200 300\n100 -1 0 4 9 9 200 300\n"
in
outputs "slice algorithms" "programs/slices.flan" slices_out;
outputs ~opt:"-O0" "slice algorithms, -O0" "programs/slices.flan" slices_out;
(* The byte predicates, parse-i64, and the two number helpers. The refused
parse-i64 cases are every shape strtoll answers 0 for — "", "abc",
"12x", "-", " 1" — so a None there is the whole reason the function is
Flan and not the bytes->i64 primitive. The RNG lines pin the actual
sequence off a fixed seed rather than just a range, which is the only
way a later change to the derivation gets caught; rand-u32 itself is
pinned by the sand hash. *)
let text_out =
"tfft\ntfftt\ntfftt\n\
1 -1 -1\n\
0 42 -42 7 9007199254740993\n\
-999 -999 -999 -999 -999\n\
1 -1 0\n\
0 2.5 10 0\n\
11 14 12 14 15\n5 5\n\
0.793725 0.324519 0.0835023\n"
in
outputs "bytes, parsing and numbers" "programs/text.flan" text_out;
outputs ~opt:"-O0" "bytes, parsing and numbers, -O0" "programs/text.flan"
text_out;
(* Rounding and sqrt. Every case here is a *negative* or a half, because
those are the two places a plausible wrong version differs: a floor
written as the bare cast truncates toward zero and answers -2 for -2.5,
and a round written as (floor-f32 (+ x 0.5)) is half-up rather than
half-away and answers -2 as well. 16777216.0 is past 2^24, where the
guard rather than the cast has to produce the answer — and where the
cast it guards would be out of i32's range. sqrt is a `declare` on
libm's sqrtf; the -O0 run is the one that matters for it, because at
-O2 LLVM folds most calls into the hardware instruction and a symbol
that never has to resolve proves nothing about the link. *)
let math_out =
"2 2 2 -3 -2 -3 0 -1 \n\
3 2 3 -2 -2 -2 1 0 \n\
0 0 -0 \n\
2 3 3 -2 -3 -3 1 -1 \n\
1.67772e+07 1.67772e+07 1.67772e+07 -1.67772e+07 \n\
0 1 2 1.41421 0.5 1000 \n\
5 \n"
in
outputs "rounding and sqrt" "programs/math.flan" math_out;
outputs ~opt:"-O0" "rounding and sqrt, -O0" "programs/math.flan" math_out;
(* index-of-bytes, trim, the byte classes and parse-f64. The search cases
are the ones that separate a correct loop from a lucky one: a match
only at the end, "aab" in "aaab" (where the first byte matches twice
before the needle does), a needle longer than the haystack, which must
answer None without building a window off the end, and the empty needle
at Some 0. trim prints inside brackets so the all-whitespace answer is
visible as [] — that input is also the one that would build a reversed
slice and trap. And parse-f64's refusals are every shape strtod hands
back a plausible number for: "", "abc", "1x", ".", "1e", " 1", "1 ",
"0x10", "nan". *)
let bytes2_out =
"6 0 4 2 1 \n\
-1 -1 -1 0 0 0 \n\
[hi][hi][hi][][][a b][x][x]\n\
ttfff\n\
ttttff\n\
0 3.5 -3.5 0.25 1000 0.015 12\n\
-999 -999 -999 -999 -999 -999 -999 -999 -999 -999 -999\n\
1 0.5\n\
2.25\n"
in
outputs "substring, trim and parse-f64" "programs/bytes2.flan" bytes2_out;
outputs ~opt:"-O0" "substring, trim and parse-f64, -O0" "programs/bytes2.flan"
bytes2_out;
(* handler-bind and signal, spec-conditions.md §1 and §2: signal returns
Unit and carries on, an unhandled one is a no-op, a nested frame does
not displace the one outside it, and the stack is restored after. *)
let conditions_out = "0\n3\n23\n3\n1103\n1103\n" in
outputs "conditions" "programs/conditions.flan" conditions_out;
outputs ~opt:"-O0" "conditions, -O0" "programs/conditions.flan" conditions_out;
outputs ~dev:true "conditions, dev" "programs/conditions.flan" conditions_out;
(* restart-case and invoke-restart, §3 to §6: the transfer itself. A
fall-through with nothing handling it, a clause reached from two frames
down, the defer in between running on the way out, an inner frame
shadowing an outer one of the same name, and a handler that returns
normally still transferring nothing. At -O0 as well, because the guard
after every call is control flow the optimiser would otherwise launder;
and as a dev build, where every one of those calls goes through a cell. *)
let restarts_out = "101\n1\n-1\n2\n7\n1010\n101\n105\n-2\n" in
outputs "restarts" "programs/restarts.flan" restarts_out;
outputs ~opt:"-O0" "restarts, -O0" "programs/restarts.flan" restarts_out;
outputs ~dev:true "restarts, dev" "programs/restarts.flan" restarts_out;
(* §2's other half, which cannot be an [outputs] case because it does not
exit 0: a handler runs, returns normally, and has still not answered the
error, so the program stops and names the condition. *)
let exe = compile "programs/error.flan" in
let code, text = run exe None in
if code <> 134 || not (contains text "handler ran")
|| not (contains text "unhandled AssetMissing")
then begin
incr failures;
Printf.printf
"FAIL an unhandled error stops the program\n\
\ got: %S (exit %d)\n wanted: exit 134, naming the condition\n"
text code
end;
(* The raylib FFI, headless. GetColor, rectangle intersection and the
shapes texture need no window, so the whole boundary is exercised
without a display: a struct out of C through an out-pointer, a struct
into C through a pointer, a keyword resolved against an enum, and a
Flan string crossing as ptr+len.
Every case is asymmetric, which is the point. 0x11223344 comes back as
four separate bytes, so a Color is not the little-endian reading of the
packed integer. The intersection of (0,0,10,4) and (6,1,10,10) is
(6,1,4,3), four numbers from four different pairs of fields, so no
permutation of Rectangle survives it. And the shapes texture is stored
or replaced by 1 1 1 1 7 depending on which field is zero, which pins
Texture2D's id and format. Handing raylib a struct and reading it back
would have passed with any of those permuted — storing and returning is
symmetric. What the last case cannot pin, because nothing raylib
computes without a GL context reads them, is width, height and mipmaps
against each other.
The camera conversions are the strongest headless material here: both
are pure arithmetic over every field of a Camera2D and two Vector2s,
and both directions are asserted as absolute answers. A round trip
would not be — the inverse cancels a permuted layout exactly, the same
way store-and-return does. The rotated pair is the only thing in the
package that pins Vector2's own two fields, because every
component-wise formula is merely mirrored by exchanging x and y and so
compares equal; a rotation mixes them. It reports ok/bad rather than a
number because sinf and cosf make the answer 27.9999981, and this
table compares stdout byte for byte. *)
let raylib_out =
"17\n34\n51\n68\n\
6\n1\n4\n3\n\
7\n13\n17\n2\n4\n3.5\n7.25\n11.5\n13.75\n\
1\n1\n1\n1\n7\n\
7\n0\n17\n2\n4\n\
28\n24\n140\n90\n\
rotated screen-to-world ok\n\
rotated world-to-screen ok\n\
point in rect yes\npoint below rect no\n\
rects overlap yes\nrects apart no\n\
circles touch yes\ncircles clear no\n\
3\n7\n\
circle meets rect yes\ncircle clears rect no\n\
circle meets line yes\ncircle clears line no\n\
point in circle yes\npoint outside circle no\n\
point in triangle yes\npoint outside triangle no\n\
point on line yes\npoint off line no\n\
point in poly yes\npoint outside poly no\n\
in square, four corners yes\nout of triangle, three no\n\
no crossing\n"
in
if Sys.command "ldconfig -p 2>/dev/null | grep -q libraylib" = 0 then begin
outputs "raylib ffi, headless" "programs/raylib-ffi.flan" raylib_out;
(* And at -O0, for the reason the rest of the table is: every struct
here crosses as (addr v) on a local, which is the alloca mem2reg
would launder before anyone noticed it was wrong. *)
outputs ~opt:"-O0" "raylib ffi, headless, -O0" "programs/raylib-ffi.flan"
raylib_out
end
else
print_endline "acceptance: skipping the raylib FFI case (no libraylib)";
(* Again at -O0. Everything above runs through mem2reg, which launders a
sloppy alloca; -O0 tests the IR actually emitted, so a disagreement
between the two points at undefined behaviour rather than a typo. *)
(* sand.flan's simulation, headless. This is the milestone-4 acceptance
case: N frames from a seeded PRNG, one hash. It imports the sim package
and not raylib, deliberately — a program that imports raylib links
libraylib on every target, and this one is the version meant to run on
wasm32 too. The hash is reproducible only because rand-f32 is ours. *)
let sand_out = "2256461126764447066\n" in
outputs "sand, headless" "programs/sand-headless.flan" sand_out;
outputs ~opt:"-O0" "sand, headless, -O0" "programs/sand-headless.flan" sand_out;
outputs ~opt:"-O0" "value semantics, -O0" "programs/values.flan" values_out;
outputs ~opt:"-O0" "machine surface, -O0" "programs/machine.flan" machine_out;
(* And once more as a dev build. Every call in one goes through a cell, so
this is the same table asserting the indirection changes nothing before
anything has been redefined — the sand hash especially, since it is the
one result that would notice a call reaching the wrong function. *)
outputs ~dev:true "sand, headless, dev" "programs/sand-headless.flan" sand_out;
outputs ~dev:true "value semantics, dev" "programs/values.flan" values_out;
outputs ~dev:true "machine surface, dev" "programs/machine.flan" machine_out;
(* Bounds checks, NEXT.md item 2. A trap has no result — it has a nonzero
exit and a message on stderr — so it needs a case shape the table above
does not have. What is asserted is the *reason*: the location, and which
index against which length. The line and column are not pinned, because
editing the program should not break the test that reads it. *)
let bounds ?opt () =
let exe = compile ?opt "programs/bounds.flan" in
let traps name arg reason =
let code, text = run exe (Some arg) in
if code <> 134
|| not (contains text "programs/bounds.flan:")
|| not (contains text reason)
then begin
incr failures;
Printf.printf
"FAIL %s\n got: %S (exit %d)\n wanted: %S (exit 134)\n"
name text code reason
end
in
(* Both edges are in bounds and must not trap: the last index of a fixed
array, a slice ending exactly at len, and an empty slice at len. *)
let code, text = run exe (Some "0") in
if text <> "0ello\n" || code <> 0 then begin
incr failures;
Printf.printf "FAIL in-bounds edges\n got: %S (exit %d)\n" text code
end;
traps "at past a fixed array" "3"
"index 3 is out of bounds for length 3";
(* Negative indices sext to a huge unsigned, so the one unsigned
comparison catches them; the message still reports the signed value. *)
traps "at with a negative index" "-1"
"index -1 is out of bounds for length 3";
traps "at past a slice" "9"
"index 9 is out of bounds for length 5";
(* A different lowering — place/Pindex, not At — so it is its own case. *)
traps "set past a fixed array" "7"
"index 7 is out of bounds for length 3";
traps "slice with hi past len" "4"
"slice [4 9) is out of bounds for length 5";
(* Without the lo <= hi test this one would not trap: it would build a
slice of length hi - lo as a huge unsigned, which is worse. *)
traps "slice with a reversed range" "2"
"slice [2 1) is out of bounds for length 5";
(try Sys.remove exe with Sys_error _ -> ())
in
bounds ();
bounds ~opt:"-O0" ();
(* The release build drops them — the calls, that is; the two declarations
stay in the header and LLVM discards the unused ones. Asserted on the IR
rather than by running an unchecked out-of-bounds program, which has no
defined behaviour to assert on. *)
let p =
Reader.read_file "programs/bounds.flan" |> Parse.program |> Check.program
in
if not (contains (Emit.program p) "call void @flan_bounds_fail(") then begin
incr failures;
print_endline "FAIL checks on: no bounds call emitted"
end;
let off = Emit.program ~checks:false p in
if contains off "call void @flan_bounds_fail(" || contains off "call void @flan_slice_fail(" then begin
incr failures;
print_endline "FAIL --no-bounds-checks: a check survived"
end;
if !failures = 0 then print_endline "acceptance: all tests passed"
else begin
Printf.printf "\n%d failure(s)\n" !failures;
exit 1
end
| _ -> print_endline "acceptance: skipped (no clang on PATH)"