gc: add `-gc boehm_leak` for leak detection (#9464)

.github/workflows/ci.yml

@@ -140,6 +140,21 @@ jobs:
       - name: Self tests with `-gc boehm`
         ## The test cases are run with non-gc `v` for now
         run: ./v -gc boehm -silent test-self
+      - name: Test leak detector
+        run: |
+          ./v -gc boehm_leak -o testcase_leak vlib/v/tests/testcase_leak.v
+          ./testcase_leak 2>leaks.txt
+          grep "Found 1 leaked object" leaks.txt && grep ", sz=1000," leaks.txt
+      - name: Test leak detector not being active for `-gc boehm`
+        run: |
+          ./v -gc boehm -o testcase_leak vlib/v/tests/testcase_leak.v
+          ./testcase_leak 2>leaks.txt
+          [ "$(stat -c %s leaks.txt)" = "0" ]
+      - name: Test leak detector not being active for normal compile
+        run: |
+          ./v -o testcase_leak vlib/v/tests/testcase_leak.v
+          ./testcase_leak 2>leaks.txt
+          [ "$(stat -c %s leaks.txt)" = "0" ]
 
   misc-tooling:
     runs-on: ubuntu-20.04

cmd/v/help/build-c.txt

@@ -72,11 +72,12 @@ see also `v help build`.
       
    -gc <mode>
       Use and link an optional garbage collector. 
-      Only `-gc boehm` is supported currently. You need to install a
+      Only `-gc boehm` and `-gc boehm_leak` are supported currently.
-      `libgc-dev` package first, or install it manually from source: 
+      You need to install a `libgc-dev` package first, or install it manually
+      from source: 
           https://github.com/ivmai/bdwgc
           
-      Note, this option is complementary to -autofree. The Boehm garbage
+      Note, `-gc boehm` is complementary to -autofree. The Boehm garbage
       collector is conservative, and it may make your program significantly
       slower if it does many small allocations in a loop. This option
       is intended *mainly* for reducing the memory usage of programs, that
@@ -84,6 +85,11 @@ see also `v help build`.
       environments like small VPSes, and for which a few ms of garbage
       collection pauses from time to time *do not matter much*.
 
+      The option `-gc boehm_leak` is intended for leak detection in
+      manual memory management. The function `gc_check_leaks()`
+      can be called to get detection results. This function is a no-op
+      when `-gc boehm_leak` is not supplied.
+
 # Miscellaneous:
    -printfn <fn_name>
       Print the content of the generated C function named fn_name.

vlib/builtin/builtin.c.v

@@ -315,9 +315,14 @@ pub fn free(ptr voidptr) {
 		return
 	}
 	$if gcboehm ? {
-		// It is better to leave it to Boehm's gc to free things.
+		// It is generally better to leave it to Boehm's gc to free things.
 		// Calling C.GC_FREE(ptr) was tried initially, but does not work
 		// well with programs that do manual management themselves.
+		//
+		// The exception is doing leak detection for manual memory management:
+		$if gcboehm_leak ? {
+			C.GC_FREE(ptr)
+		}
 		return
 	}
 	C.free(ptr)

vlib/builtin/builtin_d_gcboehm.v

@@ -9,18 +9,38 @@ $if windows {
 $if macos {
 	#pkgconfig bdw-gc
 }
-
+$if gcboehm_leak ? {
+	#define GC_DEBUG
+}
 #include <gc.h>
 
 #flag -lgc
 
+// replacements for `malloc()/calloc()`, `realloc()` and `free()`
+// for use with Boehm-GC
+// Do not use them manually. They are automatically chosen when
+// compiled with `-gc boehm` or `-gc boehm_leak`.
 fn C.GC_MALLOC(n size_t) voidptr
 
 fn C.GC_REALLOC(ptr voidptr, n size_t) voidptr
 
 fn C.GC_FREE(ptr voidptr)
 
-fn C.GC_set_find_leak(int)
+// explicitely perform garbage collection now! Garbage collections
-
+// are done automatically when needed, so this function is hardly needed 
-// fn C.CHECK_LEAKS()
 fn C.GC_gcollect()
+
+// functions to temporarily suspend/resume garbage collection
+fn C.GC_disable()
+
+fn C.GC_enable()
+
+// returns non-zero if GC is disabled
+fn C.GC_is_disabled() int
+
+// for leak detection it is advisable to do explicit garbage collections
+pub fn gc_check_leaks() {
+	$if gcboehm_leak ? {
+		C.GC_gcollect()
+	}
+}

vlib/builtin/builtin_notd_gcboehm.v

@@ -10,5 +10,7 @@ fn C.GC_REALLOC(ptr voidptr, n size_t) voidptr
 
 fn C.GC_FREE(ptr voidptr)
 
-// fn C.CHECK_LEAKS()
+// provide an empty function when manual memory management is used
-fn C.GC_gcollect()
+// to simplify leak detection
+//
+pub fn gc_check_leaks() {}

vlib/v/gen/c/cgen.v

@@ -423,7 +423,7 @@ pub fn (mut g Gen) init() {
 		}
 		g.comptime_defines.writeln('')
 	}
-	if g.pref.gc_mode == .boehm {
+	if g.pref.gc_mode in [.boehm, .boehm_leak] {
 		g.comptime_defines.writeln('#define _VGCBOEHM (1)')
 	}
 	if g.pref.is_debug || 'debug' in g.pref.compile_defines {

vlib/v/gen/c/cmain.v

@@ -67,8 +67,11 @@ fn (mut g Gen) gen_c_main_function_header() {
 
 fn (mut g Gen) gen_c_main_header() {
 	g.gen_c_main_function_header()
-	if g.pref.gc_mode == .boehm {
+	if g.pref.gc_mode in [.boehm, .boehm_leak] {
 		g.writeln('#if defined(_VGCBOEHM)')
+		if g.pref.gc_mode == .boehm_leak {
+			g.writeln('\tGC_set_find_leak(1);')
+		}
 		g.writeln('\tGC_INIT();')
 		g.writeln('#endif')
 	}
@@ -155,8 +158,11 @@ pub fn (mut g Gen) gen_c_main_for_tests() {
 	main_fn_start_pos := g.out.len
 	g.writeln('')
 	g.gen_c_main_function_header()
-	if g.pref.gc_mode == .boehm {
+	if g.pref.gc_mode in [.boehm, .boehm_leak] {
 		g.writeln('#if defined(_VGCBOEHM)')
+		if g.pref.gc_mode == .boehm_leak {
+			g.writeln('\tGC_set_find_leak(1);')
+		}
 		g.writeln('\tGC_INIT();')
 		g.writeln('#endif')
 	}

vlib/v/pref/pref.v

@@ -20,6 +20,7 @@ pub enum BuildMode {
 pub enum GarbageCollectionMode {
 	no_gc
 	boehm
+	boehm_leak
 }
 
 pub enum OutputMode {
@@ -158,7 +159,7 @@ pub mut:
 	build_options       []string // list of options, that should be passed down to `build-module`, if needed for -usecache
 	cache_manager       vcache.CacheManager
 	is_help             bool // -h, -help or --help was passed
-	gc_mode             GarbageCollectionMode = .no_gc // .no_gc, .boehm
+	gc_mode             GarbageCollectionMode = .no_gc // .no_gc, .boehm, .boehm_leak
 	// checker settings:
 	checker_match_exhaustive_cutoff_limit int = 10
 }
@@ -230,8 +231,13 @@ pub fn parse_args(known_external_commands []string, args []string) (&Preferences
 						res.gc_mode = .boehm
 						parse_define(mut res, 'gcboehm')
 					}
+					'boehm_leak' {
+						res.gc_mode = .boehm_leak
+						parse_define(mut res, 'gcboehm')
+						parse_define(mut res, 'gcboehm_leak')
+					}
 					else {
-						eprintln('unknown garbage collection mode, only `-gc boehm` is supported')
+						eprintln('unknown garbage collection mode, only `-gc boehm` and `-gc boehm_leak` are supported')
 						exit(1)
 					}
 				}

vlib/v/tests/testcase_leak.v

@@ -0,0 +1,12 @@
+// This program is supposed to test the leak detector which
+// is integrated with `-gc boehm_leak` at compile time.
+//
+// If compiled with `-gc boehm` or without `-gc` nothing
+// will be reported.
+
+fn main() {
+	mut y := unsafe { malloc(1000) }
+	// unsafe { free(y) } // leak if commented out
+	y = voidptr(0)
+	gc_check_leaks()
+}