Add http_(get|clear)_last_reponse_headers() functions

[Girgias]

This is to provide an alternative to the magic variable $http_response_header

[nielsdos]

@Girgias I think this patch should fix the crashes:

diff --git a/ext/standard/basic_functions.c b/ext/standard/basic_functions.c
index 60d484341a..bd5efb384c 100644
--- a/ext/standard/basic_functions.c
+++ b/ext/standard/basic_functions.c
@@ -423,6 +423,8 @@ PHP_RINIT_FUNCTION(basic) /* {{{ */
 	BASIC_RINIT_SUBMODULE(dir)
 	BASIC_RINIT_SUBMODULE(url_scanner_ex)
 
+	ZVAL_UNDEF(&BG(last_http_headers));
+
 	/* Setup default context */
 	FG(default_context) = NULL;
 
@@ -489,6 +491,8 @@ PHP_RSHUTDOWN_FUNCTION(basic) /* {{{ */
 	BASIC_RSHUTDOWN_SUBMODULE(user_filters)
 	BASIC_RSHUTDOWN_SUBMODULE(browscap)
 
+	zval_ptr_dtor(&BG(last_http_headers));
+
 	BG(page_uid) = -1;
 	BG(page_gid) = -1;
 	return SUCCESS;
diff --git a/ext/standard/http.c b/ext/standard/http.c
index 6276fdb2d1..921172a305 100644
--- a/ext/standard/http.c
+++ b/ext/standard/http.c
@@ -242,6 +242,6 @@ PHP_FUNCTION(http_last_response_header)
 	zend_parse_parameters_none();
 
 	if (!Z_ISUNDEF(BG(last_http_headers))) {
-		ZVAL_DUP(return_value, &BG(last_http_headers));
+		ZVAL_COPY(return_value, &BG(last_http_headers));
 	}
 }

The DUP->COPY change is only a performance improvement. It is not necessary to fix the crashes.

[Girgias]

Currently away from home, will try this on Monday :)

[nielsdos]

Code looks correct, and I personally think it's a sensible feature.
But since Jakub is the code owner here I'll let him have the final say.

[bukka]

I think this whole thing is an anti-pattern so we should not introduce a special function for that

[bukka]

What I would prefer to see is a function that give headers from the stream resource. Something like stream_get_headers...

[Girgias]

@bukka any final remarks before I merge this?

[bukka]

LGTM

[nielsdos]

You should check the return value, or use the macro version I think. Same comment for further below.

[Girgias]

Indeed, and I didn't initially what you meant .-. so now I'm also using explicit return macros.

[nielsdos]

When using RETURN_COPY_VALUE, you should not pass return_value.

[Girgias]

Yeah that was a hastly change that I didn't test locally :/

[nielsdos]

Strictly speaking the engine initializes it to NULL for you, so this is not necessary but eh... better be explicit ig.

[nielsdos]

Actually this function returns void so RETURN_NULL is confusing here.

[nielsdos]

Actually this function returns void so RETURN_NULL is confusing here.

[nielsdos]

I believe it is correct now, let's see what CI says