Hi Vladimir,
The cast to void on the read is exactly how one is supposed to indicate
that the result *should* be ignored. MHD_itc_clear() is only supposed to
clear the pipe, and if it failed (say SIGINT), it'll be retried anyway,
so no harm there, the code is correct and should IMO not be changed.
Same for the fgets(): it's just to terminate the benchmark, we don't
care if the user terminated with CTRL-D, CTRL-C or ENTER or typed stuff
before hitting ENTER. So ignoring it is totally fine.
I don't see a reason to change any of this.
Happy hacking!
Christian
On 10/2/26 5:21 AM, Vladimir Yavorsky wrote:
Hello GNU libmicrohttpd team,
My name is Vladimir Yavorsky, and I am a software engineer. While
building your libmicrohttpd packages, I encountered warnings about
ignored read() and fgets() results.
The following code is present in source versions 1.0.1, 1.0.2, 1.0.5,
1.0.6, 1.0.7, 1.0.8, 1.0.9 and 1.0.10:
* src/microhttpd/mhd_itc.h: the eventfd implementation of
MHD_itc_clear_() discards the read() result with a cast to void.
* src/tools/perf_replies.c: the Enter-to-stop path discards the
fgets() result.
The warnings were reproduced with unmodified 1.0.10 and GCC 13.3.0,
using these commands from the source directory:
CFLAGS='-O2 -g -D_FORTIFY_SOURCE=2 -Wunused-result' \
./configure --disable-https --disable-doc --disable-examples \
--disable-static --enable-shared
make -j2
Is ignoring these results intentional? Would you consider retrying
read() on EINTR (interruption by a signal), and distinguishing EOF (end
of input) from an input error in perf_replies, while calling
MHD_stop_daemon() correctly?
This report concerns build warnings. An HTTP server crash or remote
exploitation has not been established. Since these code paths are also
present in earlier versions, this should not be described as a new
regression in 1.0.10.
Thank you.
Best regards,
Vladimir Yavorsky
Software Engineer
