Re: [PATCH v10 11/20] index-helper: use watchman to avoid refreshing index with lstat()

[Ramsay Jones <ramsay@ramsayjones.plus.com>]

On 12/05/16 21:20, David Turner wrote:
> From: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
[snip]

>  
> +/* in ms */
> +#define WATCHMAN_TIMEOUT 1000
> +
> +static int poke_and_wait_for_reply(int fd)
> +{
> +	struct strbuf buf = STRBUF_INIT;
> +	int ret = -1;
> +	struct pollfd pollfd;
> +	int bytes_read;
> +	char reply_buf[4096];
> +	const char *requested_capabilities = "";
> +
> +#ifdef USE_WATCHMAN
> +	requested_capabilities = "watchman";
> +#endif
> +
> +	if (fd < 0)
> +		return -1;
> +
> +	strbuf_addf(&buf, "poke %d %s", getpid(), requested_capabilities);
> +	if (packet_write_gently(fd, buf.buf, buf.len))

This is not causing a problem or bug, but is none the less not
correct - as you know, packet_write_gently() takes a 'printf' like
variable argument list. (So, buf.buf is the format specifier and
buf.len is an unused arg).

I think I would write this as:

	strbuf_addf(&buf, "poke %d", getpid());
	if (requested_capabilities && *requested_capabilities)
		strbuf_addf(&buf, " %s", requested_capabilities);
	if (packet_write_gently(fd, "%s", buf.buf))

... or something similar. [Note, just typing into my email client, so
it's not been close to a compiler.]

> +		return -1;
> +	if (packet_flush_gently(fd))
> +		return -1;

Why are you sending a flush packet - doesn't the index-helper
simply ignore it?

I haven't tried this yet BTW, just reading patches as they float
on past... ;-)

ATB,
Ramsay Jones