Editor: Free up transient memory leak in do_blocks().

There has been a memory inefficiency inside of `do_blocks()` where it
parses the given content then iterates through each top-level block to
render it. Unfortunately each top-level block can considerably add to
the overall memory use involved, and once moving on to the next
top-level block, all of the allocated memory will be retained until the
end of the call to `do_blocks()`.

In this change, each parsed block sub-tree is freed via reset to null
after it has been rendered. All top-level blocks are rendered
independently of each other and so this operation is safe — there are no
data dependencies between them.

This commit also includes follow-up changes committed in [60400] and [60402].

Rewieved by audrasjb.
Merges [60316], [60400] and [60402] to the 6.8 branch.
Props dmsnell, joemcgill.
Fixes #63588.



git-svn-id: https://develop.svn.wordpress.org/branches/6.8@60435 602fd350-edb4-49c9-b593-d223f7449a82

Author: audrasjb

src/wp-includes/blocks.php

@@ -2404,11 +2404,32 @@ function parse_blocks( $content ) {
  * @return string Updated post content.
  */
 function do_blocks( $content ) {
-	$blocks = parse_blocks( $content );
-	$output = '';
+	$blocks                = parse_blocks( $content );
+	$top_level_block_count = count( $blocks );
+	$output                = '';
 
-	foreach ( $blocks as $block ) {
-		$output .= render_block( $block );
+	/**
+	 * Parsed blocks consist of a list of top-level blocks. Those top-level
+	 * blocks may themselves contain nested inner blocks. However, every
+	 * top-level block is rendered independently, meaning there are no data
+	 * dependencies between them.
+	 *
+	 * Ideally, therefore, the parser would only need to parse one complete
+	 * top-level block at a time, render it, and move on. Unfortunately, this
+	 * is not possible with {@see \parse_blocks()} because it must parse the
+	 * entire given document at once.
+	 *
+	 * While the current implementation prevents this optimization, it’s still
+	 * possible to reduce the peak memory use when calls to `render_block()`
+	 * on those top-level blocks are memory-heavy (which many of them are).
+	 * By setting each parsed block to `NULL` after rendering it, any memory
+	 * allocated during the render will be freed and reused for the next block.
+	 * Before making this change, that memory was retained and would lead to
+	 * out-of-memory crashes for certain posts that now run with this change.
+	 */
+	for ( $i = 0; $i < $top_level_block_count; $i++ ) {
+		$output      .= render_block( $blocks[ $i ] );
+		$blocks[ $i ] = null;
 	}
 
 	// If there are blocks in this content, we shouldn't run wpautop() on it later.