(MSYS2/MinGW/GCC 14) nonblock.c: In function 'curlx_nonblock': no non-blocking method was found/used/set

[LigH-de]

I did this

Compiling in media-autobuild suite for Win32 target

  CC       libcurl_la-nonblock.lo
../../lib/nonblock.c: In function 'curlx_nonblock':
../../lib/nonblock.c:82:4: error: #error "no non-blocking method was found/used/set"
   82 | #  error "no non-blocking method was found/used/set"
      |    ^~~~~
make[2]: *** [Makefile:2831: libcurl_la-nonblock.lo] Error 1

logs.zip

HAVE_IOCTLSOCKET_FIONBIO should be defined as 1 in lib/config-win32.h

I expected the following

successful compilation

curl/libcurl version

HEAD

operating system

MSYS2/MinGW

[bagder]

How do you build curl?

[LigH-de]

I start the media-autobuild_suite.bat and wait...

Details are in the attached logs.zip file (several logs and configurations collected by M-AB-S). Head of the make log:

CPPFLAGS:  -DGNUTLS_INTERNAL_BUILD -DNGHTTP2_STATICLIB -DPSL_STATIC
CFLAGS: -D_FORTIFY_SOURCE=2 -fstack-protector-strong -mtune=generic -O2 -pipe -D__USE_MINGW_ANSI_STDIO=1 -mthreads
CXXFLAGS: -D_FORTIFY_SOURCE=2 -fstack-protector-strong -mtune=generic -O2 -pipe -D__USE_MINGW_ANSI_STDIO=1
LDFLAGS: -D_FORTIFY_SOURCE=2 -fstack-protector-strong -mtune=generic -O2 -pipe -D__USE_MINGW_ANSI_STDIO=1 -static-libgcc -static-libstdc++
make -j4 

Configuration:
$ ../configure --prefix=/local32 --disable-shared --enable-static --bindir=/local32/bin-global --with-nghttp2 --with-openssl --without-gnutls --without-mbedtls --without-libssh2 --without-random --without-ca-bundle --without-ca-path --without-librtmp --with-brotli --enable-sspi --disable-debug

[vszakats]

Normally for mingw-w64, HAVE_IOCTLSOCKET_FIONBIO is defined by autotools via the generated curl_config.h.
Since it isn't, it suggests that the auto-detection fails for some unrelated reason, maybe due to the extra passed options.
Peeking into config.log could give the answer.

[LigH-de]

Yesterday and the day before it still compiled...

[vszakats]

Our tests are silent about this. It's the line checking if ioctl FIONBIO might be used... yes in the configure output.

Is there any clue in your config.log about this particular check?

[LigH-de]

https://github.com/curl/curl/files/15273796/logs.zip contains all the log files.

configure:41533: checking if ioctlsocket FIONBIO is compilable
configure:41553: ccache gcc -c -D_FORTIFY_SOURCE=2 -fstack-protector-strong -mtune=generic -O2 -pipe -D__USE_MINGW_ANSI_STDIO=1 -mthreads -Werror-implicit-function-declaration -Wno-system-headers -DGNUTLS_INTERNAL_BUILD -DNGHTTP2_STATICLIB -DPSL_STATIC -IG:/MABS/msys64/mingw32/include -IG:/MABS/msys64/mingw32/include  -IG:/MABS/msys64/mingw32/include -IG:/MABS/local32/include -IG:/MABS/msys64/mingw32/include conftest.c >&5
conftest.c: In function 'main':
conftest.c:152:41: error: passing argument 3 of 'ioctlsocket' from incompatible pointer type [-Wincompatible-pointer-types]
  152 |         if(0 != ioctlsocket(0, FIONBIO, &flags))
      |                                         ^~~~~~
      |                                         |
      |                                         int *
In file included from conftest.c:144:
G:/MABS/msys64/mingw32/include/winsock2.h:1007:76: note: expected 'u_long *' {aka 'long unsigned int *'} but argument is of type 'int *'
 1007 |   WINSOCK_API_LINKAGE int WSAAPI ioctlsocket(SOCKET s,__LONG32 cmd,u_long *argp);
      |                                                                    ~~~~~~~~^~~~
...
configure:41562: result: no
configure:41585: checking if ioctlsocket FIONBIO might be used
configure:41596: result: no

[vszakats]

Thanks. It's a legit complaint from the compiler, but this check has been like this since forever.

Would this be explained by a compiler upgrade? gcc 14 just happened.

[LigH-de]

Indeed possible...

gcc.exe (Rev2, Built by MSYS2 project) 14.1.0

[vszakats]

Would this patch fix the build for you?:
4bab28f

[LigH-de]

I have only little experience in editing the shell scripts used by the suite.
Are you able to provide this change as diff patch? Maybe github can do that if I just know how to build the matching URL...

[vszakats]

This link gives you the raw .diff:
https://github.com/curl/curl/commit/4bab28f0661d95bb6421d9ae3630cb1cca2fc487.diff

[LigH-de]

Yes, I discovered how to add this patch and curl builds again.

[vszakats]

@LigH-de Thanks for testing, and reporting! The fix is merged now.

[mickmack1213]

Sorry for reviving this. I just wanted to point this line out. I think a similar issue might happen at some point, as the wrong pointer type is passed (long instead of int though)

[vszakats]

@mickmack1213 Can you paste which line do you mean or give a permalink?

[LigH-de]

this line

This actually is a permalink, it highlights the matching line 3253 in the source view in a reasonably modern webbrowser.

  if test "$curl_cv_func_ioctlsocket_camel" = "yes"; then
    AC_MSG_CHECKING([if IoctlSocket FIONBIO is compilable])
    AC_COMPILE_IFELSE([
      AC_LANG_PROGRAM([[
        $curl_includes_bsdsocket
      ]],[[    # <===
        long flags = 0;
        if(0 != IoctlSocket(0, FIONBIO, &flags))
          return 1;
      ]])
    ],[
      AC_MSG_RESULT([yes])
      tst_compi_ioctlsocket_camel_fionbio="yes"
    ],[
      AC_MSG_RESULT([no])
      tst_compi_ioctlsocket_camel_fionbio="no"
    ])
  fi

But I believe it refers to the following two lines, not the one with the double square brackets.

[vszakats]

Ah okay, thanks for the snippet.

(In my browser, your link points to ]],[[. This will change if anything is pushed to master and wasn't clear what it meant. By permalink, I mean one that contains a commit hash to avoid ambiguity: https://docs.github.com/en/get-started/writing-on-github/working-with-advanced-formatting/creating-a-permanent-link-to-a-code-snippet)

[vszakats]

Do you have a documentation link for IoctlSocket()?

[LigH-de]

The official Microsoft Winsock2 documentation specifies the flag (or iMode in the example there) as u_long (assuming that means unsigned long).

[vszakats]

Are you sure this snippet is for Windows? the casing of the function looks different.

[LigH-de]

No, I only remembered that ioctlsocket is one of the functions in Winsock[2]. But now I noticed that the presence of $curl_includes_bsdsocket slipped me; and BSD Sockets are quite the same.

But I confess, I do not know where exactly the M4 script looks for it.