Issue 10344 Exiting _Dmain should flush all FILE*s and return nonzero on failure

[lionello]

Flushing all open streams on exit, throwing an Error with the errno string in case of failure.

TODO: test cases. But must get test suite to work first.

[andralex]

I suggest we do this instead of the above:

extern (C) int main(int argc, char **argv)
{
    auto result = _d_run_main(argc, argv, &_Dmain);
    if (fflush(stdout) != 0)
    {
        fprintf(stderr, "Failed to flush stdout, error code: %d\n", errno); // unchecked
        result = 1;
    }
    fflush(stderr); // unchecked
    return result;
}

The justification goes as follows:

1. If we flush all buffers, we'll try to flush whatever stray files the user has forgotten opened. If the user didn't care to close them, we can't assume they are of consequence to the program.
2. stdout, however, is the runtime's responsibility because we're assume to take care of properly closing it.
3. Not being able to flush stderr should not change the exit code because most programs do not depend on a proper stderr.

[WalterBright]

There's a misunderstanding here. When the C runtime gets control back from main(), it calls exit(status) which flushes all open stdio buffers. Success or failure of that does not affect the exit status.

D does not need to flush and then ignore the results of the flush - that already happens. See 7.19.3 in the C99 Standard: "If the main
function returns to its original caller, or if the exit function is called, all open files are closed (hence all output streams are flushed) before program termination. Other paths to program termination, such as calling the abort function, need not close all files
properly."

The issue we have is not that the files are not flushed - they are. The issue is what to do if the flushing failed. In C stdio, failure of the flush upon program exit is ignored.

[andralex]

Indeed the flush of stderr is redundant, but other than that the code is true to form.

[andralex]

So let me reiterate the code I propose:

extern (C) int main(int argc, char **argv)
{
    auto result = _d_run_main(argc, argv, &_Dmain);
    if (fflush(stdout) != 0)
    {
        fprintf(stderr, "Failed to flush stdout, error code: %d\n", errno); // unchecked
        result = 1;
    }
    return result;
}

Is that at the appropriate time (i.e. all flushes etc will happen after main as far as I understand)?

[WalterBright]

Yes, it is at the right time, but I would amend the setting of result to be:

if (result == 0) result = 1;

[lionello]

Indeed, ignoring failures for stderr seems to be the behavior of other tools I've tried (cat asdf 2</dev/null and the like), but I have a slight preference for "throw" since it can be debugged and it'll encourage the author to add proper handling to the code. And even though the exception can't be caught from D main(), I can imagine there being a higher level (OS level?) exception handler hook capability that will get invoked.

The down side is that the stack trace might be a bit overwhelming to the enduser (when there's no debugger attached), though the human readable error message would still be there.

Also: what about rt_term? Should we fflush stdout there? (In fact, the code in rt_term is nearly identical to the one in _d_run_main and I wonder why we don't use rt_term to begin with?)

[lionello]

I've clearly done something wrong here. I tried to rebase my fix on upstream/master, but now it looks like my PR pulls in all these other changes. Closing.

[andralex]

OK, please do follow up with another diff. Thanks!