Could you add a unit test?

Indeed, the refactoring in 4cba2f9 looks fishy. If len == 1 we would access h->header[-1]. It seems to me that we could stick with the do-loop, if we change the indexes from len-1 to len in the while-clause. @weltling Am I missing something?

Regarding the unit test: I'm not sure if we can use the --HEADERS-- section since the docs say it would only be supported by server-tests.php, but sapi/tests/test006.phpt and sapi/tests/test007.phpt use it.

Regarding to do ... while, possible solution is:

if (NULL != p) {
	len = p - h->header + 1;
}
if (len > 0) {
	do {
		len--;
	} while (len != 0 && (h->header[len-1] == ' ' || h->header[len-1] == '\t'));

It works fine even if : is first or last character (which doesn't make sense).

@cmb69 If we use h->header[len] in condition, then correct solution is:

if (NULL != p) {
	len = p - h->header;
}
if (len > 0) {
	do {
		len--;
	} while (len != 0 && (h->header[len] == ' ' || h->header[len] == '\t'));
	if (len) {
		len++; // Include last character.

If len == 0 then condition is false without evaluation of h->header[len-1] (regardless to type of loop). So, I think that's better to use while loop because we should keep original value of len (if whitespace doesn't exist) and it's little faster (and more clear solution for me).

I've added test and we don't need --HEADERS-- because we should use header function to send header.

@stodorovic Thanks, that sounds plausible. (And yes, I've confused apache_response_headers with apache_request_headers).

Comment on behalf of cmb at php.net:

Thanks! Applied via 47b89bc.