On Wed, 19 Jan 2022 09:20:07 GMT, Xin Liu <x...@openjdk.org> wrote: >>> This is in the signal handler. Is it safe to print message or do something >>> complex? >> >> raw write to stdout should be fine I think. >> >>>I would not print anything here, first because it is in the signal handler >>>and secondly I don't think it is needed. No real thread dump usages are >>>going to hit a VM that hasn't initialized yet IMO. >> >> Might be nice for the user to know why his jcmd or kill -3 gets ignored >> though. > > because JVM_HANDLE_XXX_SIGNAL is installed in os::init_2(), it's safe to use > log. I added a log entry with the tag os+init. -Xlog:os+init=info will print > out a message "ignore BREAK_SIGNAL in the initialization phase." I also > update the testing example in the description section.
Sorry for being difficult, but I would not do UL logging here. UL is heavyweight, and a shifting target too, because folks like to extend it with new capabilities. God knows what log_info really does (initialization, network traffic, accessing thread locals...) or will do in the future. I'd keep it simple stupid and either just use write to fd=2 or fprintf(stderr). The former is safe to use in signal handling contexts, not sure about the latter. Otherwise maybe David is right and we skip the output altogether. ------------- PR: https://git.openjdk.java.net/jdk/pull/7003
