torture test fails in 951

[icing]

I did this

Running torture test 951 on master:

********* System characteristics ********
* curl 8.19.0-DEV (x86_64-apple-darwin24.6.0)
* libcurl/8.19.0-DEV OpenSSL/3.5.4 zlib/1.2.12 brotli/1.1.0 zstd/1.5.7 libidn2/2.3.8 libpsl/0.21.5 libssh2/1.11.1_DEV nghttp2/1.67.1 ngtcp2/1.17.0 nghttp3/1.12.0 librtmp/2.3 mit-krb5/1.7-prerelease OpenLDAP/2.6.10
* Protocols: dict file ftp ftps gopher gophers http https imap imaps ipfs ipns ldap ldaps mqtt mqtts pop3 pop3s rtmp rtsp scp sftp smtp smtps telnet tftp ws wss
* Features: alt-svc AppleSecTrust AsynchDNS brotli Debug GSS-API HSTS HTTP2 HTTP3 HTTPS-proxy HTTPSRR IDN IPv6 Kerberos Largefile libz PSL SPNEGO SSL SSLS-EXPORT threadsafe UnixSockets zstd
* Disabled: win32-ca-searchpath, win32-ca-search-safe, override-dns
* Host: resistance.fritz.box
* System: Darwin resistance.fritz.box 24.6.0 Darwin Kernel Version 24.6.0: Wed Nov  5 21:30:23 PST 2025; root:xnu-11417.140.69.705.2~1/RELEASE_X86_64 x86_64
* OS: darwin
* Perl: v5.34.1 (perl)
* Args: -t 951
* Env: nghttpx-h3
* Seed: 240522
* Servers: SSL HTTP-IPv6 HTTP-unix FTP-IPv6
*****************************************
test 0951...[SMTP data with dot as first character]
 114 functions to make fail
** MEMORY FAILURE
Leak detected: memory still allocated: 64 bytes
At 600002748328, there is 64 bytes.
 allocated by curl_addrinfo.c:137
LIMIT hostip.c:584 calloc reached memlimit
 951: torture FAILED: function number 88 in test.
 invoke with "-t88" to repeat this single case.

 - abort tests
TESTDONE: 1 tests were considered during 10 seconds.
TESTDONE: 0 tests out of 1 reported OK: 0%

FAIL 951: 'SMTP data with dot as first character' SMTP

TESTFAIL: These test cases failed: 951

I expected the following

test passing

curl/libcurl version

master

operating system

macOS

[vszakats]

Could reproduce. With -DENABLE_THREADED_RESOLVER=OFF (--disable-threaded-resolver) the leak goes away.

[vszakats]

This issue predates the ability to run torture tests with the threaded resolver enabled, finalized via fb7033d.

[bagder]

I can't seem to reproduce on Linux!

[bagder]

Does libbacktrace work on macOS? Can you reproduce using a build with that enabled and then attach the tests/log/memdump file for the torture fail?

[vszakats]

curl builds and links fine with libbacktrace (both Apple clang and llvm/clang), but their memdump look like this:

MEM curl/lib/getenv.c:67 strdup(0x16b82e563) (2) = 0x6000026c7548
MEM curl/lib/escape.c:191 free(0x6000026c7548)
MEM curl/src/tool_cfgable.c:39 calloc(1,1192) = 0x15800ce08
LIMIT curl/lib/easy.c:196 malloc reached memlimit
compile with -g
compile with -g
compile with -g
compile with -g
compile with -g
compile with -g
compile with -g
LIMIT curl/src/tool_paramhlp.c:37 calloc reached memlimit
compile with -g                                                              
compile with -g                                                              
compile with -g                                                              
compile with -g                                                              
compile with -g                                                              
compile with -g                                                                               
compile with -g                                                                               
compile with -g                                                                               
compile with -g                                                                               
compile with -g
MEM curl/src/tool_cfgable.c:202 free(0x15800ce08)

(source is compiled with -g, as confirmed by verbose log)

When compiled with gcc-15 with libbacktrace enabled, memdump and test output looks identical to a non-libbacktrace-enabled build.

In theory libbacktrace should recognize mach-o DWARF.
I tried applying suggestions from ElementsProject/lightning#8431 via ianlancetaylor/libbacktrace#158, which made the result the same as with gcc-15, i.e. no backtraces :(

edit: how does libbacktrace output look like when working correctly?

[bagder]

edit: how does libbacktrace output look like when working correctly?

It provides a stack trace. Source file name, line number and function name all the way from the function until the main() call.

[bagder]

Here's what the memdump file looks like on my Linux machine after I ran runtests.pl -n -t13 1

memdump-stacktrace.txt

[icing]

Fix in #20468

[vszakats]

I confirm #20468 fixes the test failure.