On Sat, Apr 01, 2023 at 04:20:25PM +0200, Omar Polo wrote:
> On 2023/04/01 15:26:53 +0200, Christian Weisgerber <naddy@mips.inka.de> wrote:
> > When I try to rebase jca@'s jca/llvm15-12 branch on the latest
> > OpenBSD master, I correctly get a merge conflict.  I logged the
> > output of the whole rebase operation:
> > 
> > $ got rb llvm15 2>&1 | tee ../log
> > 
> > That produces some 11,200 lines.  Some 200 lines from the end,
> > splatted into another line of output, there's a "got: conflicts
> > must be resolved before rebasing can continue".  It's also thousands
> > of lines from the actual C conflict line.
> > 
> > Looking at cmd_rebase(), I see that "conflicts must be resolved
> > before rebasing can continue" should appear at the very end.  It's
> > issued to stderr, so written right away, while the other output
> > goes to stdout and is stuck in a buffer that is only written out
> > once got exits.  In short, cmd_rebase() or one of the functions it
> > calls is missing a fflush(stdout)... but I don't know where the
> > most logical place for that would be.
> 
> it could be an idea to flush stdout before printing the error message
> in main() to cover all the (sub)commands

This cannot hurt, can it? Tests are passing. I'm ok with this.

> diff /home/op/w/got
> commit - e9e0377f452e9d3f600011e0714cc6c779f10bab
> path + /home/op/w/got
> blob - a51a450a06ab2800d2fe4a200fc31cc59364dac6
> file + got/got.c
> --- got/got.c
> +++ got/got.c
> @@ -265,6 +265,7 @@ main(int argc, char *argv[])
>  		      error->code == GOT_ERR_ERRNO && errno == EPIPE) &&
>  		    !(sigint_received &&
>  		      error->code == GOT_ERR_ERRNO && errno == EINTR)) {
> +			fflush(stdout);
>  			fprintf(stderr, "%s: %s\n", getprogname(), error->msg);
>  			return 1;
>  		}
> 
>