cgen: move closure code to vlib #24912

kbkpbot · 2025-07-16T11:56:09Z

This PR:

move some closure codes from cgen to vlib/builtin.
add a benchmark of closure in vlib/bench/bench_closure.v

TODO:
freestanding can't work now because lack of pthread_mutex

huly-for-github · 2025-07-16T11:56:23Z

Connected to Huly®: V_0.6-23381

kbkpbot · 2025-07-16T11:58:21Z

benchmark result:

before this PR:

V Language Closure Performance Benchmark Report

1. Closure Performance Analysis

Test Name	Iterations	Time(ms)	Ops/sec	Notes
Normal Function Call	100000000	0	+inf Mop/s	Baseline
---------------------------	------------	----------	--------------	--------------
Small Closure Creation	10000000	392	25.51 Mop/s
Medium Closure Creation	10000000	662	15.11 Mop/s
Large Closure Creation	1000000	221	4.52 Mop/s
---------------------------	------------	----------	--------------	--------------
Small Closure Call	100000000	264	378.79 Mop/s
Medium Closure Call	100000000	246	406.50 Mop/s
Large Closure Call	10000000	26	384.62 Mop/s
---------------------------	------------	----------	--------------	--------------
Multi-threaded Creation	1000000	228	4.39 Mop/s	Threads: 8 Iterations per thread: 125000

2. Memory Overhead Analysis

Test Name	Closure Count	Start Mem(KB)	End Mem(KB)	Delta(KB)	Bytes/Closure
Closure Memory Overhead	100000	`1076224`	`1083008`	6784	69

Verification Sum: -900 (Calculated from random sample of 100 closures)

Test Environment

V Language Version: V 0.4.11 8605599
CPU Cores: 8
System Memory: 15884 MB
Operating System: linux

Test Time: 2025-07-16 19:07:52.807113

After this PR:

V Language Closure Performance Benchmark Report

1. Closure Performance Analysis

Test Name	Iterations	Time(ms)	Ops/sec	Notes
Normal Function Call	100000000	0	+inf Mop/s	Baseline
---------------------------	------------	----------	--------------	--------------
Small Closure Creation	10000000	413	24.21 Mop/s
Medium Closure Creation	10000000	674	14.84 Mop/s
Large Closure Creation	1000000	222	4.50 Mop/s
---------------------------	------------	----------	--------------	--------------
Small Closure Call	100000000	236	423.73 Mop/s
Medium Closure Call	100000000	246	406.50 Mop/s
Large Closure Call	10000000	24	416.67 Mop/s
---------------------------	------------	----------	--------------	--------------
Multi-threaded Creation	1000000	177	5.65 Mop/s	Threads: 8 Iterations per thread: 125000

2. Memory Overhead Analysis

Test Name	Closure Count	Start Mem(KB)	End Mem(KB)	Delta(KB)	Bytes/Closure
Closure Memory Overhead	100000	1075968	1082880	6912	70

Verification Sum: -900 (Calculated from random sample of 100 closures)

Test Environment

V Language Version: V 0.4.11 a11b69e
CPU Cores: 8
System Memory: 15884 MB
Operating System: linux

Test Time: 2025-07-16 19:57:45.160076

spytheman · 2025-07-16T13:35:16Z

vlib/builtin/closure.v

I am against this being directly in builtin, since it will affect even programs that do not use closures at all.

It could be in a builtin.closure module instead, that gets imported selectively, similar to how sync is imported for channels.

It will also solve the problem of more exotic environments breaking (like freestanding etc).

By moving it in builtin.closure, I encounter some errors:

FAIL [ 35/2449] C: 1512.3 ms, R: 0.000 ms vlib/builtin/js/array_test.js.v >> compilation failed: builder error: error: import cycle detected between the following modules: * builtin -> builtin.closure -> builtin * main -> builtin -> builtin.closure -> builtin * builtin.closure -> builtin -> builtin.closure * os -> builtin -> builtin.closure -> builtin * strings.textscanner -> builtin -> builtin.closure -> builtin FAIL [ 36/2449] C: 1420.1 ms, R: 0.000 ms vlib/builtin/js/string_test.js.v >> compilation failed: builder error: error: import cycle detected between the following modules: * builtin -> builtin.closure -> builtin * main -> builtin -> builtin.closure -> builtin * builtin.closure -> builtin -> builtin.closure * os -> builtin -> builtin.closure -> builtin * strings.textscanner -> builtin -> builtin.closure -> builtin OK [ 37/2449] C: 214.7 ms, R: 1.441 ms vlib/builtin/wchar/wchar_test.v FAIL [ 38/2449] C: 1542.9 ms, R: 0.000 ms vlib/builtin/js/int_test.js.v >> compilation failed: builder error: error: import cycle detected between the following modules: * builtin -> builtin.closure -> builtin * main -> builtin -> builtin.closure -> builtin * builtin.closure -> builtin -> builtin.closure * os -> builtin -> builtin.closure -> builtin * strings.textscanner -> builtin -> builtin.closure -> builtin OK [ 39/2449] C: 244.9 ms, R: 1.559 ms vlib/cli/command_test.v

debugging.......

try adding builtin.closure in builtin_module_parts in vlib/v/util/util.v:17

although that may need adding that change separately in a single PR, so that the compiler can bootstrap later with it on master

spytheman · 2025-07-16T13:48:55Z

Interesting ... why is there a diff for the Small Closure Call case?

vlib/builtin/closure/closure.v

vlib/builtin/closure/closure.c.v

spytheman · 2025-07-27T13:19:11Z

vlib/v/parser/expr.v

+				if p.file_backend_mode == .v || p.file_backend_mode == .c {
+					p.register_auto_import('builtin.closure')
+				}


not all anonymous functions are closures ... many of them do not capture anything

we should probably make this more specific in the future 🤔

spytheman · 2025-07-27T13:22:08Z

vlib/v/parser/parser.v

+		if p.file_backend_mode == .v || p.file_backend_mode == .c {
+			p.register_auto_import('builtin.closure')
+		}


I am not sure about this either ... the parameter may be a simple anonymous function, that does not capture anything and does not need allocating thunks etc.

spytheman · 2025-07-27T13:24:59Z

vlib/v/parser/struct.v

@@ -284,6 +284,22 @@ fn (mut p Parser) struct_decl(is_anon bool) ast.StructDecl {
 					// error is set in parse_type
 					return ast.StructDecl{}
 				}
+
+				// for field_name []fn, cgen will generate closure, so detect here


why would cgen generate a closure for a field init?
afaik it only does it for stuff like field: instance.method , not for normal functions, whether anonymous or not

spytheman

Excellent work.

* scanner: fix multi-level string interpolation in if/match branch * scanner: fix multi-level string interpolation in if/match branch, fix test code * Update vlib/v/scanner/scanner.v Co-authored-by: Delyan Angelov <[email protected]> * rename the rest * remove vfmt off/on tags * cgen: fix array data for option array/fixed array(?[]u8/?[3]u8) * cgen: fix fixed array with type alias * cgen: fix fixed array with type alias * cgen: move sort fn after interface definitions * parser: fix syntax error for c in [othermod.Struct{field: 255}] { * comptime: support string interpolation in `compile warn/error` * comptime: support string interpolation in `compile warn/error` * fix error for string literal * make level less than 0 for compile warn/error * cgen: move closure C code to V code under vlib/builtin/closure/ (vlang#24912) * ci: workaround -usecache issue afte 2d87ac4 * markused,checker: fix hello world size after the introduction of `builtin.closure` in 2d87ac4 (vlang#24989) * time: fix more panics in the supported specifiers in Time.custom_format/1 (vlang#24988) --------- Co-authored-by: krchi <[email protected]> Co-authored-by: Delyan Angelov <[email protected]> Co-authored-by: kbkpbot <[email protected]>

cgen: move closure code to vlib

c3db4f6

kbkpbot added 2 commits July 16, 2025 20:48

move independ code to closure.v

b5e0086

add pthread_mutex_t def

7672b40

spytheman reviewed Jul 16, 2025

View reviewed changes

kbkpbot added 13 commits July 17, 2025 13:43

move to builtin.closure, fix js

bb97302

fix for cgen extra gen_closure_fn()

901c33f

Merge remote-tracking branch 'origin/master' into move-closure-to-vlib

3be4e77

fix array anon funcs

78f076f

Merge master into move-closure-to-vlib

ab05ce0

fix -usecache

15a1618

fix windows header

2214125

Merge branch 'vlang:master' into move-closure-to-vlib

dc59652

add vinix

e1733aa

fix vinix

989cc89

Merge branch 'vlang:master' into move-closure-to-vlib

f69c97b

Merge branch 'vlang:master' into move-closure-to-vlib

e61fd44

workaound -usecache bug

35b4a7e

spytheman reviewed Jul 26, 2025

View reviewed changes

vlib/builtin/closure/closure.v Show resolved Hide resolved

kbkpbot added 5 commits July 26, 2025 16:55

parser: only import builtin.closure when .c or .v

0719464

Merge branch 'vlang:master' into move-closure-to-vlib

38df657

add readme

131d8aa

remove duplicate cond check

31767fe

Merge branch 'vlang:master' into move-closure-to-vlib

d894d86

spytheman reviewed Jul 27, 2025

View reviewed changes

vlib/builtin/closure/closure.c.v Show resolved Hide resolved

spytheman reviewed Jul 27, 2025

View reviewed changes

spytheman approved these changes Jul 27, 2025

View reviewed changes

spytheman reviewed Jul 27, 2025

View reviewed changes

spytheman approved these changes Jul 27, 2025

View reviewed changes

spytheman merged commit 2d87ac4 into vlang:master Jul 27, 2025
80 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

cgen: move closure code to vlib #24912

cgen: move closure code to vlib #24912

kbkpbot commented Jul 16, 2025 •

edited

Loading

Uh oh!

huly-for-github bot commented Jul 16, 2025

Uh oh!

kbkpbot commented Jul 16, 2025

Uh oh!

spytheman Jul 16, 2025

Uh oh!

spytheman Jul 16, 2025

Uh oh!

kbkpbot Jul 16, 2025

Uh oh!

spytheman Jul 16, 2025

Uh oh!

spytheman Jul 16, 2025

Uh oh!

spytheman commented Jul 16, 2025

Uh oh!

Uh oh!

Uh oh!

spytheman Jul 27, 2025

Uh oh!

spytheman Jul 27, 2025

Uh oh!

spytheman Jul 27, 2025

Uh oh!

spytheman Jul 27, 2025

Uh oh!

spytheman left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cgen: move closure code to vlib #24912

cgen: move closure code to vlib #24912

Conversation

kbkpbot commented Jul 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

huly-for-github bot commented Jul 16, 2025

Uh oh!

kbkpbot commented Jul 16, 2025

V Language Closure Performance Benchmark Report

1. Closure Performance Analysis

2. Memory Overhead Analysis

Test Environment

V Language Closure Performance Benchmark Report

1. Closure Performance Analysis

2. Memory Overhead Analysis

Test Environment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

spytheman commented Jul 16, 2025

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

spytheman left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

kbkpbot commented Jul 16, 2025 •

edited

Loading