From: Junio C Hamano <gitster@pobox.com>
Date: Sun, 14 Jun 2020 18:24:10 GMT
Subject: Re: [PATCH] pkt-line: extract out PACKET_HEADER_SIZE
Message-ID: <xmqqtuzdy08l.fsf@gitster.c.googlers.com>
In-Reply-To: <xmqqimfuzz73.fsf@gitster.c.googlers.com>

Junio C Hamano <gitster@pobox.com> writes:

>> How about (this ugly code):
>>
>> 	packet_trace("0000", sizeof "0000" - 1, 1);
>> 	if (write_in_full(fd, "0000", sizeof "0000" - 1) < 0)
>
> Yeah, that is ugly.  I was thinking more in the direction of
> replacing these three-argument write_in_full with something like
>
> #define write_constant(fd, constant_string) \
> 	write_in_full((fd), (constant_string), strlen(constant_string))
>
> with some preprocessor magic to make the compilation break when the
> second parameter to the macro is not a string constant.

There is a bit of subtlety but I did mean C preprocessor macro and
not a helper function with the above.  With use of a macro defined
like above, the programmer can write

	write_constant(fd, "0000");

which would turn into

	write_in_full((fd), "0000", strlen("0000"));
	
Descent compilers know to produce identical code as

	write_in_full((fd), "0000", 4);

when seeing a literal constant string given to strlen().

But a helper function like this:

	static int write_constant(int fd, const char *string) {
		return write_in_full(fd, string, strlen(string));
	}

has less chance of getting the same kind of optimization (the helper
needs to be inlined before the compiler can realize that the
parameter to strlen() is a literal constant whose length can be
computed at the compile time).