remove warnings in mdfour.c, checksum.h, trace.h, netint.c #165
Conversation
|
Thanks for sending this pull request and highlighting the warnings. In retrospect, one of these warnings appeared in the travis test logs. I'll make a point of actually looking at these logs and not just checking that the tests passed in the future. Interestingly on my system a newer version of gcc also gives a warning for netint.c; one that requires a surprising level of optimization insight to notice. It's noticed that for the following line "d" will not have been initialized if rs_suck_netint() doesn't return RS_DONE; Line 143 in d50d5c9 I think gcc also gives a slightly faster Release build rdiff binary than clang does. I"m going to play with various ways to make the select's less warn-y and maybe faster and get back to this. |
Experimenting here with commiting directly to someone else's pull request in github. This is the better way to fix this.
gcc version 9.2.1 20190821 (Debian 9.2.1-4) gave a warning that 'd' is not always initialized by rs_suck_netint(). It noticed that if rs_suck_netint() doesn't return RS_DONE it means it doesn't set the value. This stops that warning.
Tests show this is faster for gcc than the switch statements, as well as avoiding warnings.
|
I'm happy with this now. After the tests run and the logs show no warnings I'm going to merge. Thanks for your contribution. |
This removes warnings for
mdfour.c(using unsigned long long to avoid casting warning on Windows due to different data models),checksum.h(give a return value of 0 after the switch statement to avoidcontrol reaches end of non-void functionwarning), andtrace.h(setHAVE_PRINTF_Zto off when cross-compiling for Windows inCMakeLists.txt)