fix: fix zmalloc_size on macos

[andydunstall]

Fixes hset_family_test tests on MacOS

Looks like zmalloc_size isn't defined properly on Mac, so zmalloc uses src/redis/zmalloc_mi.c definition but zmalloc_size uses malloc_size

void *zp = zmalloc(10);
cout << "zmalloc; zmalloc_size " << zmalloc_size(zp) << endl;  // 0
cout << "zmalloc; malloc_size " << malloc_size(zp) << endl;  // 0
cout << "zmalloc; zmalloc_usable_size " << zmalloc_usable_size(zp) << endl;  // 16
void *mp = malloc(10);
cout << "malloc; malloc_size " << malloc_size(mp) << endl;  // 16
cout << "malloc; zmalloc_size " << zmalloc_size(mp) << endl;  // 16

This just copies the USE_ZMALLOC_MI branch from libc into macos branch (I don't really understand this allocator stuff but with this the hset_family_test tests pass running locally on Mac - will run rest of the tests on mac)

[romange]

Added some of my fixes as well. Proof that it is passing unit tests: https://github.com/dragonflydb/dragonfly/actions/runs/8006667920

[kostasrim]

Hi Andy, thank you for fixing this!

I suppose you have a mac? Can I sincerely ask you to try something for me. I would like to understand why this doesn't fail on Redis, since they use the exact same header on macos. My assumption is that Redis uses jemalloc so they actually call je_malloc_usable_size(p). If you do have a mac and you are happy to try out you can add an #error to check out which path gets exercised in the header. If it does exercises the #elif defined(__APPLE__) instead this is something that we should probably take a look at.

I pre approve but I would like this to be investigated a little more

p.s. I rerun a daily job to see the failure on the CI as well

[romange]

They do not have exactly the same header. We use mi_malloc, they do not. I changed this header to fit our requirements

…

On Fri, Feb 23, 2024 at 12:12 PM Kostas Kyrimis ***@***.***> wrote: ***@***.**** approved this pull request. Hi Andy, thank you for fixing this! I suppose you have a mac? Can I sincerely ask you to try something for me. I would like to understand why this doesn't fail on Redis, since they use the exact same header on macos. My assumption is that Redis uses jemalloc so they actually call je_malloc_usable_size(p). If you do have a mac and you are happy to try out you can add an #error to check out which path gets exercised in the header. If it does exercises the #elif defined(__APPLE__) instead this is something that we should probably take a look at. I pre approve but I would like this to be investigated a little more — Reply to this email directly, view it on GitHub <#2646 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA4BFCBPZHJUAEIIRPUGWSDYVBTRRAVCNFSM6AAAAABDVBCQKCVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMYTQOJXG43TQMBQGI> . You are receiving this because you commented.Message ID: ***@***.***>

-- Roman Gershman CTO --- *www.dragonflydb.io <http://www.dragonflydb.io>*

[kostasrim]

They do not have exactly the same header. We use mi_malloc, they do not. I changed this header to fit our requirements

I see, cheers!