On Wed, May 17, 2023 at 10:37:24AM +0200, Omar Polo wrote:
> On 2023/05/17 10:34:36 +0200, Omar Polo <op@omarpolo.com> wrote:
> > we're pretty loud when reporting write failures in fcgi socket (and
> > what these causes later, i.e. various GOT_ERR_CANCELLED.)  While we
> > have to handle the disconnection and stop doing work, I don't really
> > see the point in logging these by default.  A client that disconnects
> > earlier is not that of a big deal.
> > 
> > So, diff below turns the log_warn("write failure") into log_debug, so
> > you can still see them when running with -v, and don't log the
> > GOT_ERR_CANCELLED that we use to signal our callers to stop.
> > 
> > ok?
> 
> Sorry, wrong diff attached, was a previous attempt that didn't kept
> the return -1 in got_output_blob_by_lines().

> @@ -901,8 +902,9 @@ got_output_blob_by_lines(struct template *tp, struct g
>  	free(line);
>  
>  	if (err) {
> -		log_warnx("%s: got_object_blob_getline failed: %s",
> -		    __func__, err->msg);
> +		if (err->code == GOT_ERR_CANCELLED)

Should't the above line say != instead of == ?

> +			log_warnx("%s: got_object_blob_getline failed: %s",
> +			    __func__, err->msg);
>  		return -1;
>  	}
>  	return 0;