GC Memory leak
al1-ce opened this issue · 10 comments
OS info
OS: Arch Linux
LDC: v1.37.0
Steps to reproduce
touch test.d
echo "void main() {}" >> test.d
ldc test.d
valgrind --leak-check=full --show-leak-kinds=all ./test
Valgrind trace
Short (no --leak-check=full)
==40149== Memcheck, a memory error detector
==40149== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.
==40149== Using Valgrind-3.22.0 and LibVEX; rerun with -h for copyright info
==40149== Command: ./test
==40149==
==40149==
==40149== HEAP SUMMARY:
==40149== in use at exit: 124 bytes in 4 blocks
==40149== total heap usage: 1,085 allocs, 1,081 frees, 524,365 bytes allocated
==40149==
==40149== LEAK SUMMARY:
==40149== definitely lost: 0 bytes in 0 blocks
==40149== indirectly lost: 0 bytes in 0 blocks
==40149== possibly lost: 32 bytes in 1 blocks
==40149== still reachable: 92 bytes in 3 blocks
==40149== suppressed: 0 bytes in 0 blocks
==40149== Rerun with --leak-check=full to see details of leaked memory
==40149==
==40149== For lists of detected and suppressed errors, rerun with: -s
==40149== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
Full (click me)
==39912== Memcheck, a memory error detector
==39912== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.
==39912== Using Valgrind-3.22.0 and LibVEX; rerun with -h for copyright info
==39912== Command: ./test
==39912==
==39912==
==39912== HEAP SUMMARY:
==39912== in use at exit: 124 bytes in 4 blocks
==39912== total heap usage: 1,085 allocs, 1,081 frees, 524,365 bytes allocated
==39912==
==39912== 10 bytes in 1 blocks are still reachable in loss record 1 of 4
==39912== at 0x4843788: malloc (vg_replace_malloc.c:442)
==39912== by 0x4F4CF0E: strdup (strdup.c:42)
==39912== by 0x4D6D73C: rt.trace._sharedStaticCtor_L23_C1() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6AA37: rt.minfo.rt_moduleCtor().__foreachbody1(ref rt.sections_elf_shared.DSO) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6BA28: rt.sections_elf_shared.DSO.opApply(scope int(ref rt.sections_elf_shared.DSO) delegate) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D62EC8: rt_init (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6346C: rt.dmain2._d_run_main2(char[][], ulong, extern(C) int(char[][]) function).runAll() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D632C6: _d_run_main2 (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6311C: _d_run_main (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x109181: main (in /g/test)
==39912==
==39912== 10 bytes in 1 blocks are still reachable in loss record 2 of 4
==39912== at 0x4843788: malloc (vg_replace_malloc.c:442)
==39912== by 0x4F4CF0E: strdup (strdup.c:42)
==39912== by 0x4D6D75A: rt.trace._sharedStaticCtor_L23_C1() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6AA37: rt.minfo.rt_moduleCtor().__foreachbody1(ref rt.sections_elf_shared.DSO) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6BA28: rt.sections_elf_shared.DSO.opApply(scope int(ref rt.sections_elf_shared.DSO) delegate) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D62EC8: rt_init (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6346C: rt.dmain2._d_run_main2(char[][], ulong, extern(C) int(char[][]) function).runAll() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D632C6: _d_run_main2 (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6311C: _d_run_main (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x109181: main (in /g/test)
==39912==
==39912== 32 bytes in 1 blocks are possibly lost in loss record 3 of 4
==39912== at 0x4843788: malloc (vg_replace_malloc.c:442)
==39912== by 0x4D3B7CA: core.internal.gc.impl.conservative.gc.initialize() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D2CAB9: core.gc.registry.createGCInstance(immutable(char)[]) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D48B9C: gc_init (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D48C15: gc_init_nothrow (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D483E6: _DThn16_4core8internal2gc4impl5protoQo7ProtoGC6qallocMFNbmkMxC8TypeInfoZSQCn6memory8BlkInfo_ (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D48DDB: gc_qalloc (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D4F917: core.lifetime._d_newitemT!(rt.critical_.D_CRITICAL_SECTION)._d_newitemT() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D62D1D: _d_criticalenter2 (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6E41F: rt.trace._staticDtor_L408_C1() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6AAF5: rt.minfo.rt_moduleTlsDtor().__foreachbody1(ref rt.sections_elf_shared.DSO) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6BAC1: rt.sections_elf_shared.DSO.opApplyReverse(scope int(ref rt.sections_elf_shared.DSO) delegate) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912==
==39912== 72 bytes in 1 blocks are still reachable in loss record 4 of 4
==39912== at 0x484ABC0: realloc (vg_replace_malloc.c:1690)
==39912== by 0x4D35B50: core.internal.container.common.xrealloc(void*, ulong) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D35511: core.internal.container.array.Array!(core.gc.gcinterface.Range).Array.insertBack!().insertBack(core.gc.gcinterface.Range) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D485FC: _DThn16_4core8internal2gc4impl5protoQo7ProtoGC8addRangeMFNbNiPvmxC8TypeInfoZv (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D695CD: rt.memory.initStaticDataGC().__foreachbody1(ref rt.sections_elf_shared.DSO) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6BA28: rt.sections_elf_shared.DSO.opApply(scope int(ref rt.sections_elf_shared.DSO) delegate) (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D62EC3: rt_init (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6346C: rt.dmain2._d_run_main2(char[][], ulong, extern(C) int(char[][]) function).runAll() (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D632C6: _d_run_main2 (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x4D6311C: _d_run_main (in /usr/lib/libdruntime-ldc-shared.so.107.1)
==39912== by 0x109181: main (in /g/test)
==39912==
==39912== LEAK SUMMARY:
==39912== definitely lost: 0 bytes in 0 blocks
==39912== indirectly lost: 0 bytes in 0 blocks
==39912== possibly lost: 32 bytes in 1 blocks
==39912== still reachable: 92 bytes in 3 blocks
==39912== suppressed: 0 bytes in 0 blocks
==39912==
==39912== For lists of detected and suppressed errors, rerun with: -s
==39912== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
If to use -betterC switch then leak disappears
echo "extern(C) int main() { return 0; }" > test.d
ldc test.d -betterC
valgrind ./test
Outputs:
==41353== Memcheck, a memory error detector
==41353== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.
==41353== Using Valgrind-3.22.0 and LibVEX; rerun with -h for copyright info
==41353== Command: ./test
==41353==
==41353==
==41353== HEAP SUMMARY:
==41353== in use at exit: 0 bytes in 0 blocks
==41353== total heap usage: 0 allocs, 0 frees, 0 bytes allocated
==41353==
==41353== All heap blocks were freed -- no leaks are possible
==41353==
==41353== For lists of detected and suppressed errors, rerun with: -s
==41353== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
and sanitizers? (address, memory, thread)
Havent tested, not sure how to though
Havent tested, not sure how to though
Try:
$> export ASAN_OPTIONS=abort_on_error=0:fast_unwind_on_malloc=0:detect_leaks=1 UBSAN_OPTIONS=print_stacktrace=1
$> ldc2 hello.d -fsanitize=memory
$> ./hello
## Output empty/none or stacktrce?
Reference
Nothing
But running resulting binary with valgrind kills my pc
isn't leaking intentional in DMD and LDC without -lowmem
?
i misunderstood, never mind.
These are probably just harmless little leaks in upstream druntime, where nobody bothered cleaning up a few bytes. E.g., for
==39912== 32 bytes in 1 blocks are possibly lost in loss record 3 of 4
==39912== at 0x4843788: malloc (vg_replace_malloc.c:442)
==39912== by 0x4D3B7CA: core.internal.gc.impl.conservative.gc.initialize() (in /usr/lib/libdruntime-ldc-shared.so.107.1
See https://github.com/dlang/dmd/blob/274eec89f4ce7f34a1776e3ac1b1e9c82de90811/druntime/src/core/internal/gc/impl/conservative/gc.d#L142 - that ConservativeGC
object itself is malloc'd during runtime/GC initialization and most likely never freed.
Does this leak occur on another system and/or libc?
Because on the Alpine I didn't have the same problem.
I have no way of checking other platforms
Also not sure about libc, like, what I do with it?
Also not sure about libc, like, what I do with it?
I cited libc because Alpine uses musl and not glibc.
It would be interesting to know if this applies only to glibc, or even only to archlinux.
If you have docker installed, you can test it:
docker-ldc2
# ldc2 master - upstream (musl fixed)
$ docker pull kassany/alpine-ldc2
$ docker run --rm -it -v $(pwd):/app -w /app kassany/alpine-ldc2:latest ash
# for musl can build `ldc2/ldmd2 -static foo.d` (libunwind-static already installed)
# run valgrind (your host) in static-foo.