Make check errors on v0.8.2rc2

[grubles]

Issue and Steps to Reproduce

Running $ make check results in this error.

[user@box lightning]$ make full-check
lightningd: WARNING: default network changing in 2020: please set network=testnet in config!
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all bitcoin/test/run-tx-encode > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all bitcoin/test/run-bitcoin_block_from_hex > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all bitcoin/test/run-secret_eq_consttime > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-key_derive > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-sphinx > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-softref > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-cryptomsg > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-funding_tx > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-bolt11 > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-gossip_rcvd_filter > /dev/null
VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-bigsize > /dev/null
==30055== Invalid read of size 4
==30055==    at 0x10027E88: main (run-bigsize.c:295)
==30055==  Address 0x477bd4e is 9,262 bytes inside a block of size 9,265 alloc'd
==30055==    at 0x40842AC: malloc (vg_replace_malloc.c:309)
==30055==    by 0x10021E57: allocate (tal.c:250)
==30055==    by 0x10022B5F: tal_alloc_ (tal.c:428)
==30055==    by 0x10022C97: tal_alloc_arr_ (tal.c:471)
==30055==    by 0x10023A27: tal_dup_ (tal.c:799)
==30055==    by 0x10020F0B: tal_strdup_ (str.c:18)
==30055==    by 0x100213B7: tal_strsplit_ (str.c:156)
==30055==    by 0x10027CBB: main (run-bigsize.c:289)
==30055== 
make: *** [Makefile:456: unittest/common/test/run-bigsize] Error 7

OS: CentOS 8
Arch: ppc64le

note: I did not run into this on Ubuntu 18.04 (on same arch) so it's possible it's something CentOS specific.

[niftynei]

Failed to repro on a centos x86_64 box (4.18.0-147.3.1.el8_1.x86_64)

[niftynei]

per discussion with @rustyrussell, this is likely due to a bug/oversight in the PPC-Valgrind implementation. (Successfully reproduced on centos ppc64le).

In the meantime, I'd suggest running the checks with the address sanitizer enabled instead (though i'm not 100% sure it works since it's using cc not gcc?)

/.configure --disable-valgrind --enable-address-sanitizer

[mpe]

I agree this is probably a ppc64le/valgrind issue.

I can reproduce on Fedora 31 with valgrind 3.15:

VALGRIND=1 valgrind -q --error-exitcode=7 --track-origins=yes --leak-check=full --show-reachable=yes --errors-for-leak-kinds=all common/test/run-bigsize > /dev/null
==22135== Invalid read of size 4
==22135==    at 0x100278A8: main (run-bigsize.c:295)
==22135==  Address 0x477bd5e is 9,262 bytes inside a block of size 9,265 alloc'd
==22135==    at 0x40841AC: malloc (vg_replace_malloc.c:309)
==22135==    by 0x10021C67: allocate (tal.c:250)
==22135==    by 0x10022703: tal_alloc_ (tal.c:428)
==22135==    by 0x1002283B: tal_alloc_arr_ (tal.c:471)
==22135==    by 0x100235B7: tal_dup_ (tal.c:799)
==22135==    by 0x10020AAB: tal_strdup_ (str.c:18)
==22135==    by 0x10020F57: tal_strsplit_ (str.c:156)
==22135==    by 0x100277A3: main (run-bigsize.c:289)
==22135== 

[mpe]

And also with valgrind git 0788b1cf7 NEWS: Mention the MinGW include/valgrind.h fix.

[shenki]

There was a fix made for a similar (same?) issue a few years ago:

https://bugs.kde.org/show_bug.cgi?id=386945

However it indicates the issue was fixed, which @mpe has shown it is not.

[rustyrussell]

Thanks to @shenki and @mpe I've filed a bug at https://bugs.kde.org/show_bug.cgi?id=420780