Test leaks produces unrelated memory leaks

[cspiel1]

Currently it is difficult to develop unit tests because there are some unrelated memory leaks reported with their backtrace.

If this happens:

	if (mstat_after.blocks_cur > mstat_before.blocks_cur) {
		mem_debug();
		re_assert(false && "Test leaks memory blocks");
	}

memory leaks related to libre_init() are reported.

Is a libre_close() missing here?

[sreimers]

The check ensures that each individual test does not leak memory. So after each test memory usage should like before, otherwise the tests leaks memory. Which tests fails? On main branch everything looks good to me.

Edit: maybe a global cache is not closed/cleaned properly for your new test (like dns cache etc.)

[cspiel1]

Yes, I know.

Here: #1520

[cspiel1]

Reproduce with mentioned branch from #1520.

~/src/re/build $ ./test/retest -d ../test/data -r test_dns_parallel -v

[cspiel1]

Not interesting here are the mem leaks that originate from

• test/main.c:194
• test/main.c:198

[sreimers]

There are leaked DNS queries (you can ignore the unrelated re_main objects):

  0x55b607cf0fd0: nrefs=1  size=18      [74 65 73 74 31 2e 65 78 61 6d 70 6c 65 2e 6e 65 ] [test1.example.ne]
./build/test/retest(+0xc56d5) [0x55b5f85d26d5]
	btrace at /home/runner/work/re/re/src/mem/mem.c:23 (discriminator 1)
./build/test/retest(mem_alloc+0xa8) [0x55b5f85d281e]
	mem_alloc at /home/runner/work/re/re/src/mem/mem.c:169
./build/test/retest(dns_dname_decode+0x100) [0x55b5f85b0e3a]
	dns_dname_decode at /home/runner/work/re/re/src/dns/dname.c:181 (discriminator 1)
./build/test/retest(dns_rr_decode+0x6c) [0x55b5f85b1f11]
	dns_rr_decode at /home/runner/work/re/re/src/dns/rr.c:224
./build/test/retest(+0xa0f66) [0x55b5f85adf66]
	reply_recv at /home/runner/work/re/re/src/dns/client.c:299
./build/test/retest(+0xa130b) [0x55b5f85ae30b]
	udp_recv_handler at /home/runner/work/re/re/src/dns/client.c:376
./build/test/retest(+0xe8536) [0x55b5f85f5536]
	udp_read at /home/runner/work/re/re/src/udp/udp.c:207
./build/test/retest(+0xe8597) [0x55b5f85f5597]
	udp_read_handler at /home/runner/work/re/re/src/udp/udp.c:221
./build/test/retest(+0xc3128) [0x55b5f85d0128]
	fd_poll at /home/runner/work/re/re/src/main/main.c:931
./build/test/retest(re_main+0x17a) [0x55b5f85d0433]
	re_main at /home/runner/work/re/re/src/main/main.c:1065
./build/test/retest(re_main_timeout+0x7d) [0x55b5f858ead3]
	re_main_timeout at /home/runner/work/re/re/test/test.c:1048
./build/test/retest(+0x3463c) [0x55b5f854163c]
	test_dns_parallel_param at /home/runner/work/re/re/test/dns.c:701
./build/test/retest(test_dns_parallel+0x8f) [0x55b5f854184a]
	test_dns_parallel at /home/runner/work/re/re/test/dns.c:740
./build/test/retest(+0x80756) [0x55b5f858d756]
	test_unit at /home/runner/work/re/re/test/test.c:587
./build/test/retest(test_reg+0x5e) [0x55b5f858e05f]
	test_reg at /home/runner/work/re/re/test/test.c:805
./build/test/retest(main+0x390) [0x55b5f855c294]
	main at /home/runner/work/re/re/test/main.c:236

[cspiel1]

Yes, after analyzing everything else I came to the same conclusion. But it takes some effort each time a memory leak happens during development of a unit test. I opened this issues to discuss if we could limit the output to the essentials. Can we?

[cspiel1]

I tried to add a libre_close() and the output looks better. Is it correct?

	if (mstat_after.blocks_cur > mstat_before.blocks_cur) {
		libre_close();
		mem_debug();
		re_assert(false && "Test leaks memory blocks");
	}

It looks similar to how it is done in the main() function in test/main.c.

Edit: commit f1d55b6

[sreimers]

Not sure about this, if leak is related to libre (timers, file descriptors, async threads) these things are now hidden. mem_debug() prints the list by added order, so the relevant parts for test leakage are most likely at the end. re_assert is also nice for debugger usage, if you close libre all context is gone.

[cspiel1]

I'll do some experiments with this PR: #1523