[cknitt]

Why

Parsetree is the AST that PPXs see, that -bs-ast serializes, and that #8624 wants to freeze as "v1". A principle there is that the parsetree describes source code only.

Externals broke that. After the built-in PPX had read an external's FFI attributes (@send, @module, @as, @inline, …), it replaced the primitive string with the computed specification:

and primitive_repr =
  | Prim_name of string
  | Prim_ffi of {name: string; spec: External_ffi_types.t}
  | Prim_inline_const of External_ffi_types.inline_const

So parsetree.ml depended on the compiler's internal FFI representation, the type had two cases no PPX can observe, and the v0 bridge needed an error branch for them.

What changes

pval_prim is the string as written, with its location: string loc option. The attributes stay where the user put them, and the type checker interprets them, once.

For

@send external on: (t, @as("click") _, unit => unit) => unit = "addEventListener"
@inline let answer = 42

the tree after the built-in PPX (bsc -only-parse -dparsetree) changes from

value_description "on"
  core_type
    Ptyp_arrow        (* arity 2: the @as argument already erased *)
    …
  "addEventListener"
  <ffi spec>
value_description "answer"
  core_type
    Ptyp_constr "int"
  <inline const>

to

value_description "on"
  attribute "send"
  core_type
    Ptyp_arrow        (* arity 3, with the Parg_fixed "click" argument *)
    …
  "addEventListener"
value_description "answer"
  attribute "inline"
    [ Pstr_eval (Pexp_constant PConst_int (42,None)) ]
  core_type
    Ptyp_constr "int"
  "#rescript-inline"

How the pieces fit:

• Type checker resolves. Typedecl.transl_value_decl calls Primitive.resolve_external, a hook that returns the resolved type, the attributes resolution did not consume, the primitive name and its kind (Kind_intrinsic, Kind_external spec, Kind_inline_const c). It types the resolved type, so .cmi/.cmj and the generated JavaScript are what they were.
• The hook. The resolver (Ast_external.resolve) lives above compiler/ml, so it is a hook; Ppx_entry, which every compiler driver links, registers it at module initialization, like the Oprint hooks. It runs in the type checker because -bs-ast writes the parsetree after the built-in PPX and a later bsc process reads it back: the specification must be recomputable from the tree.
• Diagnostics come from the type checker. The resolver reports an external's errors and warnings, and checks unused attributes on the resolved form (Bs_ast_invariant's pass skips FFI externals). It runs inside the declaration's @warning scope, so @warning("-101") on an external now silences its unused-attribute warning. Errors and warning 101 now come in source order with type errors rather than in the parse step; the reprint path and -bs-ast alone no longer report them.
• Relative @module paths. Such an external must not be inlined in other modules, since its path is relative to the declaring file. The built-in PPX used to wrap it in include (struct external … end : {let …}). Now the type checker does the equivalent: in a signature, Typedecl gives it Val_reg; in a structure, Typemod builds that constrained include in the typed tree, whose coercion (Tcoerce_primitive) emits the exported function. Primitive.no_cross_module_inline reads relativeness off the resolved spec (External_ffi_types.is_relative, split out of check_decl), so Ast_external_process.resolution loses its no_inline_cross_module flag. The JavaScript is unchanged, and so are hover, references and definition in the editor.
• Each unused attribute is reported once. A relative @module external used to keep its resolved form as a val in the wrapper's interface, which was checked as well, so its unused attributes were reported twice.
• @inline constants become external answer: int = "#rescript-inline" with the @inline(<literal>) attribute kept; Ast_external_mk has the helpers that build and read such a declaration, and Bs_builtin_ppx shrinks to one case per item instead of one per literal kind. A structural node for them is a separate Stamp parsetree v1: what remains before the AST can be frozen for PPXs #8624 item.
• @inline let bigint constants are inlined now, like the other literals and like bigint constants in signatures. Before, a let skipped them: @inline let big = 10n stayed an exported binding, with warning 53 saying @inline cannot appear there.
• Ast_external_process.handle_attributes is resolve, returning a resolution record (pval_type, spec, pval_attributes) instead of a tuple and a second wrapper.
• Printers. Outcometree.oval_prim is the value's resolved Primitive.description, so hover and interface output still render the FFI attributes, and Primitive.print is gone. Printast, Pprintast and the ReScript printer print the string as written.
• Typed tree. Typedtree.val_prim is gone: its two readers, the typed tree printer and gentype, use the value's val_kind (Val_prim), which has the resolved primitive.
• v0 bridge. Ast_mapper_to0 no longer needs an error branch; Ast_mapper_from0 gives the string a ghost location, since v0 has none.
• Ast_attributes.rs_externals is renamed is_ffi_external and takes the primitive name. Bs_ast_invariant loses a Bs_ffi_warning branch that could not fire.

Under -bs-no-builtin-ppx, a plain-named external is now resolved as an FFI external by the type checker, where it used to be typed as an intrinsic of that name.

Cost

Each external is resolved once, in the compile step, which now reads it as written. Allocated words for the parse and compile steps together, measured locally against master:

File	master	this PR
Stdlib_Math.resi	335,281	337,421 (+0.6%)
Stdlib_Math.res	337,081	342,615 (+1.6%)
Stdlib_Console.res	391,697	397,048 (+1.4%)

The parse step allocates less than on master; the rest is the compile step reading each external's attributes as written and resolving them there.

Tests

• tests/syntax_tests/data/ast-mapping/Externals.res: intrinsic, @val, @send, relative @module, @scope, @obj, @as and @variadic externals print back unchanged after the v0 round trip.
• ounit_ast_mapper0_tests: an external's primitive string and @send attribute survive Ast_mapper_to0 and Ast_mapper_from0.
• ounit_string_literal_tests builds its inline constants with Ast_external_mk.inline_const_of_expression.
• tests/build_tests/super_errors/fixtures/warning_101_external_arg_attribute.res: an @as on a non-_ argument of a @send external still reports warning 101, once per external, in source order.
• tests/build_tests/super_errors/fixtures/warning_101_relative_module_external.res: the same on a relative @module external is reported once, where it used to be reported twice, and @warning("-101") on an external silences it.
• tests/tests/src/relative_path_use.res: another module calls relative externals (top level, nested module, functor, declared in a .resi as external or as let) through the exported values, and inlines only the non-relative one.
• tests/analysis_tests/tests/src/RelativeExternal.res: hover, references and definition on relative externals.
• tests/tests/src/inline_const.res: @inline let bigint constants, positive and negative, are inlined.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Kbf3gzUzT5G

[codecov]

Codecov Report

❌ Patch coverage is 87.97468% with 19 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.00%. Comparing base (32116b5) to head (85671c0).
⚠️ Report is 3 commits behind head on master.

Files with missing lines	Patch %	Lines
compiler/ml/printtyped.ml	0.00%	4 Missing ⚠️
compiler/frontend/ast_external_mk.ml	87.50%	2 Missing ⚠️
compiler/frontend/bs_builtin_ppx.ml	91.66%	2 Missing ⚠️
compiler/gentype/translate_signature.ml	50.00%	2 Missing ⚠️
compiler/frontend/ast_external.ml	88.88%	1 Missing ⚠️
compiler/frontend/bs_ast_invariant.ml	87.50%	1 Missing ⚠️
compiler/ml/external_ffi_types.ml	92.30%	1 Missing ⚠️
compiler/ml/primitive.ml	75.00%	1 Missing ⚠️
compiler/ml/printast.ml	0.00%	1 Missing ⚠️
compiler/syntax/src/res_outcome_printer.ml	66.66%	1 Missing ⚠️
... and 3 more

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #8728      +/-   ##
==========================================
+ Coverage   81.95%   82.00%   +0.04%     
==========================================
  Files         456      456              
  Lines       65612    65659      +47     
==========================================
+ Hits        53775    53845      +70     
+ Misses      11837    11814      -23     

Files with missing lines	Coverage Δ
compiler/core/lam_compile_external_call.ml	92.98% <ø> (ø)
compiler/frontend/ast_attributes.ml	94.20% <100.00%> (+2.42%)	⬆️
compiler/frontend/ast_exp_handle_external.ml	80.48% <100.00%> (ø)
compiler/frontend/ast_external_process.ml	77.51% <100.00%> (-0.15%)	⬇️
compiler/frontend/ast_literal.ml	100.00% <100.00%> (ø)
compiler/frontend/ppx_entry.ml	91.66% <ø> (ø)
compiler/gentype/translation.ml	92.85% <100.00%> (+1.42%)	⬆️
compiler/ml/ast_iterator.ml	94.62% <100.00%> (+0.01%)	⬆️
compiler/ml/ast_mapper.ml	98.49% <100.00%> (+<0.01%)	⬆️
compiler/ml/ast_mapper_from0.ml	79.49% <100.00%> (+0.02%)	⬆️
... and 20 more

... and 9 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[pkg-pr-new]

Open in StackBlitz

rescript

npm i https://pkg.pr.new/rescript@8728

@rescript/belt

npm i https://pkg.pr.new/@rescript/belt@8728

@rescript/darwin-arm64

npm i https://pkg.pr.new/@rescript/darwin-arm64@8728

@rescript/linux-arm64

npm i https://pkg.pr.new/@rescript/linux-arm64@8728

@rescript/linux-x64

npm i https://pkg.pr.new/@rescript/linux-x64@8728

@rescript/runtime

npm i https://pkg.pr.new/@rescript/runtime@8728

@rescript/win32-x64

npm i https://pkg.pr.new/@rescript/win32-x64@8728

commit: 85671c0

[github-actions]

Developer playground preview: https://rescript-lang.github.io/rescript/dev-playground/?version=pr-8728

[github-actions]

Changelog

🐛 Bug fix

• Report an unused attribute on a relative @module external once instead of twice.
• Inline @inline let bigint constants like other literals, instead of warning that @inline cannot appear there.

🏠 Internal

• Keep resolved FFI specifications out of the parsetree, which now holds an external's primitive string as written. Resolve FFI externals during type checking #8728

Semantic Diff

5 commits · Full page

✨ Resolve an external's FFI attributes in the type checker

_{9d69d5d · cknitt + Claude · Oct 16, 2026, 19:28 UTC · new}

Why

The parsetree is what PPXs see and what -bs-ast serializes, and #8624 wants it to describe source only. After the built-in PPX had read an external's @send, @module, @as… attributes, it replaced the primitive string with the computed FFI specification (Prim_ffi, Prim_inline_const), so Parsetree depended on External_ffi_types and carried cases no PPX can observe. Now pval_prim is the string as written, with its location, and Typedecl resolves the attributes through Primitive.resolve_external, a hook the frontend registers since the resolver lives above compiler/ml. The generated JavaScript is unchanged.

AST modified (externals after the built-in PPX)

type t
@send external on: (t, @as("click") _, unit => unit) => unit = "addEventListener"
@inline let answer = 42

expanded

 …
 structure_item
   Pstr_primitive
   value_description "on"
+    attribute  "send"
+      []
     core_type
       Ptyp_arrow
-      arity = 2
+      arity = 3
       Nolabel
       core_type
         Ptyp_constr "t"
         []
       Nolabel
+      Parg_fixed
+        value
+          Pfixed_string (source="click", semantic="click")
+      Nolabel
       core_type
         Ptyp_arrow
         …
       core_type
         Ptyp_constr "unit"
         []
     "addEventListener"
-    <ffi spec>
 structure_item
   Pstr_primitive
   value_description "answer"
+    attribute  "inline"
+        structure_item
+          Pstr_eval
+          expression
+            Pexp_constant PConst_int (42,None)
     core_type
       Ptyp_constr "int"
       []
-    <inline const>
+    "#rescript-inline"

Typedtree modified (externals)

type t
@send external on: (t, @as("click") _, unit => unit) => unit = "addEventListener"
@inline let answer = 42

typed

 …
 structure_item
   Tstr_primitive
   value_description on/1002
     core_type
       Ttyp_arrow
       Nolabel
       core_type
         Ttyp_constr "t/1001"
         []
       Nolabel
       core_type
         Ttyp_arrow
         …
       core_type
         Ttyp_constr "unit/5"
         []
     "addEventListener"
-    <ffi spec>
 structure_item
   Tstr_primitive
   value_description answer/1003
     core_type
       Ttyp_constr "int/1"
       []
-    <inline const>
+    "#rescript-inline"

Review

• Why does the frontend still read the attributes? To report errors and warnings where they were reported, to check unused attributes on the resolved form, and to learn whether a relative @module path keeps other modules from inlining the external, which wraps it in a constrained include as before. The type checker resolves again with warnings off, so nothing is reported twice.
• Why a hook rather than passing the result along? -bs-ast writes the parsetree after the built-in PPX and a later bsc process reads it back, so the specification has to be recomputable from the tree; Ast_external registers the resolver at module initialization, as the Oprint hooks are, and every entry point that type checks links it.
• What is "#rescript-inline"? The primitive of an @inline constant's declaration, which keeps its @inline(<literal>) attribute; the resolver turns it into Kind_inline_const. A structural form is a separate Stamp parsetree v1: what remains before the AST can be frozen for PPXs #8624 item.
• Do PPXs see a difference? No: they run before the built-in PPX, and the v0 bridge carries the string as before; its error branch for already-resolved externals is gone.
• What changes under -bs-no-builtin-ppx? An external with a plain name is now resolved as an FFI external by the type checker; before, it was typed as an intrinsic of that name.

✨ Check an external's unused attributes where it is resolved

_{548eab4 · cknitt + Claude · Nov 02, 2026, 19:49 UTC · new}

Why

The frontend resolved each FFI external twice, once to validate it and once more for the unused-attribute check after the built-in PPX, and the resolved val that a relative @module external keeps for its interface was checked as well, so its unused attributes were reported twice. Ast_external.validate now runs the check on the form it has just resolved, the later pass skips FFI externals, and an attribute is reported at most once. The structure mapper maps each item before the rest, so these diagnostics stay in source order.

Diagnostic modified (unused attribute on a relative @module external)

@module("./local") external relative: (@as("v") int) => unit = "relative"

diagnostic

 Warning number 101
 A.res:1:40-42
 Unused attribute: @as
-
-Warning number 101
-A.res:1:40-42
-Unused attribute: @as

Review

• Does the order of warnings change? Between externals, no: the mapper now maps items left to right (the right operand of :: was evaluated first, which had reversed them). An external's unused attributes are reported while it is mapped, so they come before the later pass's warnings for other items, such as integer overflow.

✨ Read a value's primitive from its kind, not copies of it

_{3cd7c05 · cknitt + Claude · Nov 02, 2026, 19:51 UTC · new}

Typedtree.val_prim kept the primitive string as written, but its two readers (the typed tree printer and gentype) only need the resolved primitive, which the value's val_kind already has; the field is gone. Outcometree.oval_prim is that Primitive.description too, instead of out_primitive, a copy of its kind with the name, so Primitive.print is gone as well.

✨ Inline @inline let bigint constants like other literals

_{59a5808 · cknitt + Claude · Nov 02, 2026, 19:53 UTC · new}

Why

An @inline bigint constant in a signature was inlined, but in a let the built-in PPX skipped bigints, so the binding stayed an ordinary exported value and the attribute was reported as misplaced. Both forms now read the constant the same way.

JS modified (@inline let with a bigint)

@inline let big = 10n
let use = () => big

js

 function use() {
   return 10n;
 }
-let big = 10n;
-exports.big = big;
 exports.use = use;

Diagnostic removed (@inline on a bigint let)

@inline let big = 10n

diagnostic · Before

Warning number 53
the @inline attribute cannot appear in this context

✨ Pass an FFI external's primitive name to the code that resolves it

_{d7f09dd · cknitt + Claude · Nov 02, 2026, 19:54 UTC · new}

Ast_attributes.is_ffi_external takes the primitive name rather than pval_prim, so each caller matches Some {txt = prim_name} once and hands the name on, and Ast_external.validate loses an "empty primitive string" error it could not reach. Primitive.resolve_external keeps its string argument for the same reason: Typedecl has already matched it.

[github-actions]

Changelog

🐛 Bug fix

• Report an unused attribute on a relative @module external once instead of twice.
• A @warning attribute on an external now applies to the warnings about its FFI attributes, such as an unused @as.
• Inline @inline let bigint constants like other literals, instead of warning that @inline cannot appear there.

🏠 Internal

• Keep resolved FFI specifications out of the parsetree, which now holds an external's primitive string as written. The type checker resolves each external once.

Semantic Diff

1 commit · Full page

✏️ Resolve an external's FFI attributes in the type checker

_{85671c0 · cknitt + Claude · Nov 10, 2026, 19:24 UTC · edited}

Why

The parsetree is what PPXs see and what -bs-ast serializes, and #8624 wants it to describe source only. After the built-in PPX had read an external's @send, @module, @as… attributes, it replaced the primitive string with the computed FFI specification (Prim_ffi, Prim_inline_const), so Parsetree depended on External_ffi_types and carried cases no PPX can observe. Now pval_prim is the string as written, with its location, and Typedecl resolves the attributes, once, through Primitive.resolve_external, a hook the frontend registers since the resolver lives above compiler/ml. The resolver also reports the external's errors and warnings, inside the declaration's @warning scope. The generated JavaScript is unchanged.

AST modified (externals after the built-in PPX)

type t
@send external on: (t, @as("click") _, unit => unit) => unit = "addEventListener"
@inline let answer = 42

expanded

 …
 structure_item
   Pstr_primitive
   value_description "on"
+    attribute  "send"
+      []
     core_type
       Ptyp_arrow
-      arity = 2
+      arity = 3
       Nolabel
       core_type
         Ptyp_constr "t"
         []
       Nolabel
+      Parg_fixed
+        value
+          Pfixed_string (source="click", semantic="click")
+      Nolabel
       core_type
         Ptyp_arrow
         …
       core_type
         Ptyp_constr "unit"
         []
     "addEventListener"
-    <ffi spec>
 structure_item
   Pstr_primitive
   value_description "answer"
+    attribute  "inline"
+        structure_item
+          Pstr_eval
+          expression
+            Pexp_constant PConst_int (42,None)
     core_type
       Ptyp_constr "int"
       []
-    <inline const>
+    "#rescript-inline"

AST modified (an external with a relative path after the built-in PPX)

@module("./local") external relative: int => int = "relative"

expanded

 structure_item
-  Pstr_include
-  module_expr
-    Pmod_constraint
-    module_expr
-      Pmod_structure
-        structure_item
-          Pstr_primitive
-          value_description "relative"
-            attribute  "module"
-              …
-            core_type
-              …
-            "relative"
-    module_type
-      Pmty_signature
-        signature_item
-          Psig_value
-          value_description "relative"
-            core_type
-              …
+  Pstr_primitive
+  value_description "relative"
+    attribute  "module"
+      …
+    core_type
+      …
+    "relative"

Typedtree modified (an external with a relative path)

@module("./local") external relative: int => int = "relative"

typed · Before: the constraint the built-in PPX wrote · After: the constraint Typemod adds, implicit

 structure_item
   Tstr_include
   module_expr
-    Tmod_constraint
     module_expr
       Tmod_structure
         structure_item
           Tstr_primitive
           value_description relative/1001
             …
             "relative"
-    module_type
-      Tmty_signature
-        signature_item
-          Tsig_value
-          value_description relative/1002
-            …

Typedtree modified (externals)

type t
@send external on: (t, @as("click") _, unit => unit) => unit = "addEventListener"
@inline let answer = 42

typed

 …
 structure_item
   Tstr_primitive
   value_description on/1002
     core_type
       Ttyp_arrow
       …
     "addEventListener"
-    <ffi spec>
 structure_item
   Tstr_primitive
   value_description answer/1003
     core_type
       Ttyp_constr "int/1"
       []
-    <inline const>
+    "#rescript-inline"

Diagnostic modified (unused attribute on a relative @module external)

@module("./local") external relative: @as("v") int => unit = "relative"

diagnostic

 Warning number 101
 A.res:1:39-41
 Unused attribute: @as
-
-Warning number 101
-A.res:1:39-41
-Unused attribute: @as

Diagnostic removed (unused attribute under a @warning that disables it)

@warning("-101") @module("./local")
external silenced: @as("w") int => unit = "silenced"

diagnostic · Before

Warning number 101
A.res:2:21-23
Unused attribute: @as

JS modified (@inline let with a bigint)

@inline let big = 10n
let use = () => big

js

 function use() {
   return 10n;
 }
-let big = 10n;
-exports.big = big;
 exports.use = use;

Diagnostic removed (@inline on a bigint let)

@inline let big = 10n

diagnostic · Before

Warning number 53
the @inline attribute cannot appear in this context

Review

• How does an external with a relative path stay out of other modules? Its path is relative to the file that declares it, so other modules call the value its module exports. In a signature, Typedecl gives it Val_reg. In a structure, Typemod types it as include (struct external … end : {let …}), the wrapper the built-in PPX used to write, whose coercion (Tcoerce_primitive) emits the exported function. Primitive.no_cross_module_inline reads relativeness off the resolved spec (External_ffi_types.is_relative, split out of check_decl).
• Why a hook rather than passing the result along? -bs-ast writes the parsetree after the built-in PPX and a later bsc process reads it back, so the specification has to be recomputable from the tree. Ppx_entry, which every compiler driver links, registers the resolver at module initialization, as the Oprint hooks are.
• What do diagnostics look like now? An external's errors and warning 101 come during type checking, in source order with the type errors, instead of in the parse step; an earlier type error hides them until it is fixed. The reprint path and -bs-ast alone no longer report them.
• What does this cost? Each external is resolved once, in the compile step, which reads it as written: on Stdlib_Math.resi the parse and compile steps allocate 0.6% more words than before this change (resolving in the frontend as well had made it 4.3%).
• What is "#rescript-inline"? The primitive of an @inline constant's declaration, which keeps its @inline(<literal>) attribute; the resolver turns it into Kind_inline_const. A structural form is a separate Stamp parsetree v1: what remains before the AST can be frozen for PPXs #8624 item.
• Do PPXs see a difference? No: they run before the built-in PPX, and the v0 bridge carries the string as before; its error branch for already-resolved externals is gone.
• What changes under -bs-no-builtin-ppx? An external with a plain name is now resolved as an FFI external by the type checker; before, it was typed as an intrinsic of that name.
• What else goes? Typedtree.val_prim, a copy of the primitive string whose readers only need the value's resolved kind, and Primitive.print: Outcometree.oval_prim is the Primitive.description itself.

[github-actions]

⚠️ Compiler benchmarks

Allocated words

In total: 640,767,090 → 640,963,611 (+0.03%).

Generated projects: 920,854,787 → 920,854,754 (±0.00%).

⚠️ Regression: 15 benchmarks allocate more than 1% more.

Benchmark	Base	Head	Change
packages/@rescript/runtime/Stdlib_Symbol.res	144,086	147,406	+2.30%
packages/@rescript/runtime/Stdlib_Math.res	326,831	332,053	+1.60%
packages/@rescript/runtime/Stdlib_TypedArray.res	782,659	795,150	+1.60%
packages/@rescript/runtime/Stdlib_Date.res	551,382	558,996	+1.38%
packages/@rescript/runtime/Stdlib_Console.res	369,767	374,811	+1.36%
packages/@rescript/runtime/Stdlib_Object.res	244,951	248,099	+1.29%
packages/@rescript/runtime/Stdlib_Temporal_PlainDateTime.res	436,943	442,165	+1.20%
packages/@rescript/runtime/Stdlib_RegExp.res	162,817	164,733	+1.18%
packages/@rescript/runtime/Stdlib_String.res	504,410	510,268	+1.16%
packages/@rescript/runtime/Stdlib_Temporal_ZonedDateTime.res	572,204	578,845	+1.16%

141 more changes

Benchmark	Base	Head	Change
packages/@rescript/runtime/Stdlib_Promise.res	351,719	355,728	+1.14%
packages/@rescript/runtime/Stdlib_Temporal_PlainDate.res	402,866	407,270	+1.09%
packages/@rescript/runtime/Pervasives.res	238,939	241,474	+1.06%
packages/@rescript/runtime/Stdlib_DataView.res	230,058	232,457	+1.04%
packages/@rescript/runtime/JsxEvent.res	1,088,117	1,099,058	+1.01%
packages/@rescript/runtime/Stdlib_Temporal_Duration.res	372,671	376,332	+0.98%
packages/@rescript/runtime/Stdlib_Set.res	210,614	212,428	+0.86%
packages/@rescript/runtime/Stdlib_Float.res	218,542	220,422	+0.86%
packages/@rescript/runtime/Stdlib_Intl_Locale.res	197,097	198,775	+0.85%
packages/@rescript/runtime/Stdlib_Temporal_PlainTime.res	319,801	322,483	+0.84%
packages/@rescript/runtime/Dom_storage2.res	119,493	120,493	+0.84%
packages/@rescript/runtime/Stdlib_Temporal_PlainYearMonth.res	381,509	384,680	+0.83%
packages/@rescript/runtime/Stdlib_Symbol.resi	159,849	161,151	+0.81%
packages/@rescript/runtime/Stdlib_Global.res	141,326	142,375	+0.74%
packages/@rescript/runtime/Stdlib_Temporal_Instant.res	307,209	309,434	+0.72%
packages/@rescript/runtime/Stdlib_Map.res	205,383	206,852	+0.72%
packages/@rescript/runtime/Stdlib_Dict.res	319,549	321,817	+0.71%
packages/@rescript/runtime/Stdlib_BigInt.res	206,822	208,189	+0.66%
packages/@rescript/runtime/Stdlib_BigInt.resi	186,312	187,493	+0.63%
packages/@rescript/runtime/Dom_storage.res	166,886	167,889	+0.60%
packages/@rescript/runtime/Stdlib_Math.resi	330,661	332,614	+0.59%
packages/@rescript/runtime/Stdlib_Array.res	1,293,029	1,300,460	+0.57%
packages/@rescript/runtime/Stdlib_Int.resi	205,677	206,841	+0.57%
packages/@rescript/runtime/Stdlib_Temporal_PlainMonthDay.res	268,033	269,502	+0.55%
packages/@rescript/runtime/Stdlib_Float.resi	194,403	195,451	+0.54%
packages/@rescript/runtime/Stdlib_Intl_NumberFormat.res	403,719	405,882	+0.54%
packages/@rescript/runtime/Stdlib_IteratorObject.res	250,973	252,312	+0.53%
packages/@rescript/runtime/Stdlib_Temporal_Now.res	142,907	143,667	+0.53%
packages/@rescript/runtime/Stdlib_JsError.res	230,016	231,223	+0.52%
packages/@rescript/runtime/Stdlib_WeakMap.res	134,936	135,623	+0.51%
packages/@rescript/runtime/Stdlib_Dict.resi	241,018	242,214	+0.50%
packages/@rescript/runtime/Stdlib_String.resi	500,049	502,499	+0.49%
packages/@rescript/runtime/Stdlib_RegExp.resi	188,001	188,909	+0.48%
packages/@rescript/runtime/Stdlib_JSON.resi	236,625	237,740	+0.47%
packages/@rescript/runtime/Stdlib_WeakSet.res	122,595	123,164	+0.46%
packages/@rescript/runtime/Stdlib_Intl_PluralRules.res	201,180	202,107	+0.46%
packages/@rescript/runtime/Stdlib_Date.resi	618,508	621,342	+0.46%
packages/@rescript/runtime/Stdlib_Promise.resi	357,176	358,790	+0.45%
packages/@rescript/runtime/Stdlib_Array.resi	731,792	735,033	+0.44%
packages/@rescript/runtime/Stdlib_Intl_ListFormat.res	169,175	169,869	+0.41%
packages/@rescript/runtime/Primitive_object_extern.res	114,242	114,710	+0.41%
packages/@rescript/belt/src/Belt_Int.res	165,421	166,090	+0.40%
packages/@rescript/runtime/Stdlib_JsError.resi	185,010	185,732	+0.39%
packages/@rescript/runtime/Stdlib_Console.resi	414,795	416,383	+0.38%
packages/@rescript/runtime/Stdlib_Intl_RelativeTimeFormat.res	179,816	180,492	+0.38%
packages/@rescript/runtime/Stdlib_Intl_Segmenter.res	153,921	154,488	+0.37%
packages/@rescript/belt/src/Belt_Float.res	164,752	165,354	+0.37%
packages/@rescript/runtime/Stdlib_Set.resi	229,595	230,415	+0.36%
packages/@rescript/runtime/Stdlib_Ordering.res	152,123	152,660	+0.35%
packages/@rescript/runtime/Jsx.res	121,256	121,671	+0.34%
packages/@rescript/runtime/Stdlib_Int.res	457,227	458,775	+0.34%
packages/@rescript/runtime/Stdlib_JSON.res	297,543	298,527	+0.33%
packages/@rescript/belt/src/Belt_Int.resi	149,367	149,854	+0.33%
packages/@rescript/runtime/Stdlib_ArrayBuffer.res	107,060	107,408	+0.33%
packages/@rescript/runtime/Stdlib_DataView.resi	269,445	270,319	+0.32%
packages/@rescript/runtime/Stdlib_Nullable.resi	199,638	200,285	+0.32%
packages/@rescript/belt/src/Belt_Float.resi	150,843	151,331	+0.32%
packages/@rescript/runtime/Stdlib_Global.resi	155,741	156,235	+0.32%
packages/@rescript/runtime/Stdlib_Map.resi	217,639	218,329	+0.32%
packages/@rescript/runtime/Stdlib_Intl_Collator.res	178,054	178,613	+0.31%
packages/@rescript/runtime/Primitive_array_extern.res	113,857	114,212	+0.31%
packages/@rescript/runtime/Stdlib_IteratorObject.resi	235,056	235,761	+0.30%
packages/@rescript/runtime/Stdlib_Type.res	201,437	202,008	+0.28%
packages/@rescript/runtime/Stdlib_Null.resi	204,322	204,899	+0.28%
packages/@rescript/runtime/Stdlib_Intl_Segments.res	116,121	116,448	+0.28%
packages/@rescript/belt/src/Belt_Array.resi	649,773	651,405	+0.25%
packages/@rescript/runtime/Stdlib_Intl_DateTimeFormat.res	388,726	389,663	+0.24%
packages/@rescript/runtime/Stdlib_Bool.resi	116,432	116,710	+0.24%
packages/@rescript/belt/src/Belt_MutableSetString.resi	264,513	265,127	+0.23%
packages/@rescript/belt/src/Belt_MutableSetInt.resi	264,630	265,244	+0.23%
packages/@rescript/belt/src/Belt_SetString.resi	259,453	260,023	+0.22%
packages/@rescript/belt/src/Belt_SetInt.resi	259,580	260,150	+0.22%
packages/@rescript/runtime/Primitive_dict.res	102,325	102,540	+0.21%
packages/@rescript/runtime/Stdlib_IterableIterator.resi	149,129	149,441	+0.21%
packages/@rescript/runtime/Stdlib_JsExn.resi	122,404	122,657	+0.21%
packages/@rescript/runtime/Stdlib_List.resi	409,309	410,155	+0.21%
packages/@rescript/belt/src/Belt_MutableMapString.resi	314,510	315,124	+0.20%
packages/@rescript/belt/src/Belt_MutableMapInt.resi	314,628	315,242	+0.20%
packages/@rescript/runtime/Stdlib_Float32Array.res	343,118	343,786	+0.19%
packages/@rescript/runtime/Stdlib_Float64Array.res	343,118	343,786	+0.19%
packages/@rescript/runtime/Stdlib_Int8Array.res	343,000	343,667	+0.19%
packages/@rescript/runtime/Stdlib_Uint8Array.res	343,015	343,681	+0.19%
packages/@rescript/runtime/Stdlib_Int16Array.res	343,147	343,813	+0.19%
packages/@rescript/runtime/Stdlib_BigUint64Array.res	343,254	343,920	+0.19%
packages/@rescript/runtime/Stdlib_Uint8ClampedArray.res	343,530	344,196	+0.19%
packages/@rescript/runtime/Stdlib_Uint16Array.res	343,154	343,819	+0.19%
packages/@rescript/runtime/Stdlib_Uint32Array.res	343,154	343,819	+0.19%
packages/@rescript/runtime/Stdlib_BigInt64Array.res	343,180	343,845	+0.19%
packages/@rescript/runtime/Stdlib_Int32Array.res	343,149	343,813	+0.19%
packages/@rescript/belt/src/Belt_MapString.resi	390,782	391,529	+0.19%
packages/@rescript/belt/src/Belt_MapInt.resi	390,825	391,572	+0.19%
packages/@rescript/runtime/RescriptTools.res	104,120	104,314	+0.19%
packages/@rescript/runtime/Stdlib_Generator.res	200,647	201,016	+0.18%
packages/@rescript/runtime/Stdlib_AsyncIterableIterator.resi	159,531	159,823	+0.18%
packages/@rescript/runtime/Stdlib_Option.resi	236,075	236,505	+0.18%
packages/@rescript/runtime/Primitive_exceptions.res	201,856	202,223	+0.18%
packages/@rescript/runtime/Stdlib_Generator.resi	160,178	160,469	+0.18%
packages/@rescript/belt/src/Belt_List.resi	758,030	759,396	+0.18%
packages/@rescript/runtime/Primitive_js_extern.res	239,398	239,819	+0.18%
packages/@rescript/runtime/Stdlib_AsyncGenerator.resi	165,530	165,821	+0.18%
packages/@rescript/runtime/Primitive_string_extern.res	97,857	98,028	+0.17%
packages/@rescript/runtime/Stdlib_AsyncGenerator.res	211,493	211,861	+0.17%
packages/@rescript/runtime/Stdlib_ArrayBuffer.resi	110,780	110,969	+0.17%
packages/@rescript/runtime/Stdlib_Iterator.res	235,116	235,514	+0.17%
packages/@rescript/belt/src/Belt_internalAVLset.resi	354,491	355,076	+0.17%
packages/@rescript/belt/src/Belt_HashMapString.resi	213,316	213,665	+0.16%
packages/@rescript/belt/src/Belt_HashMapInt.resi	213,431	213,779	+0.16%
packages/@rescript/runtime/Stdlib_IterableIterator.res	178,833	179,124	+0.16%
packages/@rescript/belt/src/Belt_MutableQueue.resi	206,944	207,278	+0.16%
packages/@rescript/belt/src/Belt_Option.resi	242,980	243,369	+0.16%
packages/@rescript/belt/src/Belt_MutableMap.resi	394,956	395,585	+0.16%
packages/@rescript/belt/src/Belt_HashSetString.resi	175,179	175,454	+0.16%
packages/@rescript/belt/src/Belt_HashSetInt.resi	175,292	175,567	+0.16%
packages/@rescript/belt/src/Belt_internalAVLtree.resi	452,018	452,722	+0.16%
packages/@rescript/belt/src/Belt_MutableSet.resi	397,063	397,677	+0.15%
packages/@rescript/runtime/Primitive_object.resi	116,778	116,948	+0.15%
packages/@rescript/belt/src/Belt_MapDict.resi	513,029	513,775	+0.15%
packages/@rescript/runtime/Stdlib_Iterator.resi	159,008	159,234	+0.14%
packages/@rescript/belt/src/Belt_SetDict.resi	402,431	403,001	+0.14%
packages/@rescript/belt/src/Belt_Id.resi	259,758	260,124	+0.14%
packages/@rescript/belt/src/Belt_Set.resi	436,953	437,567	+0.14%
packages/@rescript/runtime/Stdlib_AsyncIterator.resi	177,546	177,790	+0.14%
packages/@rescript/runtime/Stdlib_Result.resi	295,963	296,367	+0.14%
packages/@rescript/belt/src/Belt_Map.resi	578,948	579,736	+0.14%
packages/@rescript/runtime/Stdlib_AsyncIterator.res	295,706	296,104	+0.13%
packages/@rescript/runtime/Stdlib_Type.resi	135,551	135,727	+0.13%
packages/@rescript/belt/src/Belt_MutableStack.resi	168,936	169,153	+0.13%
packages/@rescript/runtime/Stdlib_Intl_NumberFormat_Grouping.res	159,837	160,039	+0.13%
packages/@rescript/runtime/Primitive_int_extern.res	97,539	97,660	+0.12%
packages/@rescript/runtime/Stdlib_Lazy.resi	112,105	112,244	+0.12%
packages/@rescript/belt/src/Belt_HashSet.resi	222,404	222,679	+0.12%
packages/@rescript/runtime/Stdlib_Lazy.res	222,145	222,417	+0.12%
packages/@rescript/runtime/Primitive_promise.res	96,689	96,807	+0.12%
packages/@rescript/runtime/Stdlib_Pair.res	165,629	165,830	+0.12%
packages/@rescript/runtime/Stdlib_Bool.res	146,041	146,214	+0.12%
packages/@rescript/belt/src/Belt_HashMap.resi	296,511	296,859	+0.12%
packages/@rescript/runtime/Stdlib_Intl.res	173,934	174,137	+0.12%
packages/@rescript/runtime/Stdlib_Nullable.res	331,825	332,197	+0.11%
packages/@rescript/belt/src/Belt_internalBucketsType.resi	164,493	164,675	+0.11%
packages/@rescript/belt/src/Belt_Result.resi	239,810	240,056	+0.10%
packages/@rescript/belt/src/Belt_SortArray.resi	253,595	253,855	+0.10%

Instructions

In total: 9,211,172,800 → 9,214,096,307 (+0.03%).

⚠️ Regression: 3 benchmarks execute more than 1% more instructions.

Benchmark	Base	Head	Change
packages/@rescript/runtime/Stdlib_Math.resi	21,158,495	21,430,032	+1.28%
packages/@rescript/runtime/Stdlib_TypedArray.res	41,079,388	41,593,742	+1.25%
packages/@rescript/runtime/Stdlib_Date.resi	34,102,855	34,478,711	+1.10%
packages/@rescript/runtime/Stdlib_Temporal_ZonedDateTime.res	31,030,446	31,323,533	+0.94%
packages/@rescript/runtime/Stdlib_String.resi	29,334,052	29,594,365	+0.89%
packages/@rescript/runtime/Stdlib_Temporal_PlainDateTime.res	25,401,662	25,606,474	+0.81%
packages/@rescript/runtime/Stdlib_Temporal_PlainDate.res	23,966,342	24,154,560	+0.79%
packages/@rescript/runtime/Stdlib_Array.resi	39,941,655	40,253,758	+0.78%
packages/@rescript/belt/src/Belt_Array.resi	33,675,836	33,733,263	+0.17%
packages/@rescript/runtime/Stdlib_List.resi	24,502,787	24,537,005	+0.14%

2 more changes

Benchmark	Base	Head	Change
parse/RedBlackTree.res	14,933,665	14,954,003	+0.14%
parse/RedBlackTreeNoComments.res	14,532,761	14,552,058	+0.13%

Output size

In total: 25,843,198 → 25,816,622 (-0.10%).

🎉 Improvement: 1 output is more than 1% smaller.

Benchmark	Base	Head	Change
output/tests/*.cmti	176,053	173,922	-1.21%
output/runtime/*.cmti	1,118,869	1,115,134	-0.33%
output/runtime/*.cmt	3,044,137	3,035,903	-0.27%
output/belt/*.cmti	981,298	980,059	-0.13%

Pull request 85671c0 compared with base 00e3f40 in this run.