Use output buffer and HTML tag processor to inject directives on BODY tag for full-page client-side navigation #61212
Conversation
westonruter
added
[Type] Experimental
Actually this shouldn't be surprising because it's exactly how HTML5 specifies that the BODY and HTML tags work.
While you highlight that downstream parsers may not pick up the right set of attributes for the BODY tag, every browser should, and this isn't "HTML's loose parsing rules" at play, but HTML's extremely deterministic parsing rules 😉. It's entirely reliable to add attributes to a BODY element by appending multiple BODY tags. It may not be ideal, but it's possible (this is one of the challenges with "retroactive changes" in the HTML API, because we may miss that a BODY has a given class name until we're deep into the document). |
|
The following accounts have interacted with this PR and/or linked issues. I will continue to update these lists as activity occurs. You can also manually ask me to refresh this list by adding the If you're merging code through a pull request on GitHub, copy and paste the following into the bottom of the merge commit message. To understand the WordPress project's expectations around crediting contributors, please review the Contributor Attribution page in the Core Handbook. |
There was a problem hiding this comment.
I like where this is going, but I want to point out too that we can easily test this in Core in combination with Gutenberg PRs via the Playground.
Would it be workable to start by exploring a solution in Core? I've wanted to build this new pipeline anyway and we could potentially pioneer it through this need.
Was thinking less of gutenberg_template_output_buffer and more like one final pass by WordPress that includes hooks for things like on_tag__body and on_attribute__href. I don't know ohw these would all go or perform, but in this case the code could be something more like the following…
add_action( 'template_include', static function () {
add_action( 'on_tag__body', static function ( $processor ) {
// Check if these are already set...
$processor->set_attribute( 'data-wp-…', … );
$processor->set_attribute( 'data-wp-…', … );
}, 1, 10 );
}, 0, PHP_INT_MAX );I'm not as much of a fan of actions and filters in performance-sensitive code like this; maybe there's another option to explore. Still, the idea is of hooking into HTML events instead of creating content-oriented filter and actions.
| ); | ||
| return $passthrough; | ||
| } | ||
| add_filter( 'template_include', '_gutenberg_buffer_template_output', PHP_INT_MAX ); |
There was a problem hiding this comment.
isn't an action just a filter whose return value is ignored? we could eliminate the $passthrough if we made this add_action() couldn't we?
There was a problem hiding this comment.
Unfortunately, add_action() defined as an alias:
function add_action( $hook_name, $callback, $priority = 10, $accepted_args = 1 ) {
return add_filter( $hook_name, $callback, $priority, $accepted_args );
}So if the callback doesn't return a value, then the filter is borked. I just tested:
diff --git a/lib/experimental/full-page-client-side-navigation.php b/lib/experimental/full-page-client-side-navigation.php
index 259517cfc84..452abee6c63 100644
--- a/lib/experimental/full-page-client-side-navigation.php
+++ b/lib/experimental/full-page-client-side-navigation.php
@@ -68,12 +68,8 @@ add_filter( 'gutenberg_template_output_buffer', '_gutenberg_add_client_side_navi
* } elseif ( current_user_can( 'switch_themes' ) ) {
*
* @link https://core.trac.wordpress.org/ticket/43258
- *
- * @param string $passthrough Value for the template_include filter which is passed through.
- *
- * @return string Unmodified value of $passthrough.
*/
-function _gutenberg_buffer_template_output( string $passthrough ): string {
+function _gutenberg_buffer_template_output() {
ob_start(
static function ( string $output ): string {
/**
@@ -85,6 +81,5 @@ function _gutenberg_buffer_template_output( string $passthrough ): string {
return (string) apply_filters( 'gutenberg_template_output_buffer', $output );
}
);
- return $passthrough;
}
-add_filter( 'template_include', '_gutenberg_buffer_template_output', PHP_INT_MAX );
+add_action( 'template_include', '_gutenberg_buffer_template_output', PHP_INT_MAX );The result is an empty page because the if ( $template ) condition isn't entered here: https://github.com/WordPress/wordpress-develop/blob/204a1bbf4e5f22b07a93c1f4a0b12bdd65d6483f/src/wp-includes/template-loader.php#L105-L112
There was a problem hiding this comment.
darn. you've shattered my understanding of this. I guess do_action() simply ignores the return value, but apply_filter() doesn't care if something was added as a filter or an action.
Co-authored-by: dmsnell <dmsnell@git.wordpress.org>
@dmsnell For core, I was thinking this would land in a patch like as follows: diff --git a/wp-includes/template-loader.php b/wp-includes/template-loader.php
index 0fd08545..119df0f3 100644
--- a/wp-includes/template-loader.php
+++ b/wp-includes/template-loader.php
@@ -102,6 +102,9 @@ if ( wp_using_themes() ) {
* @param string $template The path of the template to include.
*/
$template = apply_filters( 'template_include', $template );
+
+ ob_start( 'wp_template_output_buffer_callback' );
+
if ( $template ) {
include $template;
} elseif ( current_user_can( 'switch_themes' ) ) {Where function wp_template_output_buffer_callback( string $output ): string {
/**
* Filters the template output buffer prior to sending to the client.
*
* @param string $output Output buffer.
* @return string Filtered output buffer.
*/
return (string) apply_filters( 'wp_template_output_buffer', $output );
}(Aside: This also needs to account for non-HTML responses. Done in 300d9e3.) But you're saying the raw HTML output shouldn't be filterable at all, but only be accessible via the HTML Tag Processor? I will note that there are other use cases for being able to access the raw response such as via a filter which are noted in Core-43258, for example caching plugins and optimization plugins. Currently all such plugins have to reinvent the wheel to do their own output buffering of the entire response. |
westonruter
force-pushed
the
update/client-side-navigation-directive-addition
branch
from
04171bf
to
300d9e3
Compare
| return (string) $p . '<body data-wp-interactive="core/experimental" data-wp-context="{}">'; | ||
| function _gutenberg_add_client_side_navigation_directives( $response_body ) { | ||
| $is_html_content_type = false; | ||
| foreach ( headers_list() as $header ) { |
There was a problem hiding this comment.
In WordPress/performance#1189 I've improved on the testability of this by prepending 'Content-Type: ' . ini_get( 'default_mimetype' ) to this list.
No that's not it. What I was saying is that I want to find an interface that meets peoples' needs in a way that's easier, more reliable, and more performant than when everyone does their own filtering. Specifically, if we can turn 20x+ iterations over all the content into one iteration with 20x+ filters for that "chunk" or "thing" then we can potentially win a big performance improvement. For example, instead of filtering raw HTML, if Core exposes a way to filter attribute values, given the tag it's on and the content, then maybe tons of plugins would voluntarily remove their own filtering code. function add_blank_opener( $tag_name, $attribute_name, $value ) {
return `{$value} _blank`;
}
add_filter(
'html_attribute__a__target',
'add_blank_opener',
10,
3
); |
It would be great to measure the impact of running |
|
@gziolo it will be measured extensively 😉 though I predict we may not use filters directly for performance reasons. still, from what I can tell, the existing situation is still probably worse because every filter splits apart the entire document and iterates every token anyway. We’ll probably skip lots of work in the worst case |
… tag for full-page client-side navigation (WordPress#61212) * Inject client-side navigation directives via Tag Processor and output buffering * Fix typing for render_block phpdoc * Eliminate use of render_block filter * Remove copy-pasta since tag Co-authored-by: dmsnell <dmsnell@git.wordpress.org> * Ensure client-side directives are only added to HTML response bodies --------- Co-authored-by: westonruter <westonruter@git.wordpress.org> Co-authored-by: dmsnell <dmsnell@git.wordpress.org> Co-authored-by: cbravobernal <cbravobernal@git.wordpress.org>
|
@westonruter We use output buffering in our 900k+ plugin and based on my experience I suggest the following changes: <?php
function _gutenberg_buffer_template_output( string $passthrough ): string {
ob_start(
static function ( string $output, ?int $phase): string {
if ($phase & PHP_OUTPUT_HANDLER_FINAL) {
/**
* Filters the template output buffer prior to sending to the client.
*
* @param string $output Output buffer.
* @return string Filtered output buffer.
*/
return (string) apply_filters( 'gutenberg_template_output_buffer', $output );
}
return $output;
}
);
return $passthrough;
}As I remember these are related when |
… tag for full-page client-side navigation (WordPress#61212) * Inject client-side navigation directives via Tag Processor and output buffering * Fix typing for render_block phpdoc * Eliminate use of render_block filter * Remove copy-pasta since tag Co-authored-by: dmsnell <dmsnell@git.wordpress.org> * Ensure client-side directives are only added to HTML response bodies --------- Co-authored-by: westonruter <westonruter@git.wordpress.org> Co-authored-by: dmsnell <dmsnell@git.wordpress.org> Co-authored-by: cbravobernal <cbravobernal@git.wordpress.org>
|
@nextend I did quite a bit more research and I found that instead of checking for whether it is in the |
| ob_start( | ||
| static function ( string $output ): string { | ||
| /** | ||
| * Filters the template output buffer prior to sending to the client. | ||
| * | ||
| * @param string $output Output buffer. | ||
| * @return string Filtered output buffer. | ||
| */ | ||
| return (string) apply_filters( 'gutenberg_template_output_buffer', $output ); | ||
| } | ||
| ); |
There was a problem hiding this comment.
Based on WordPress/performance#1317, I believe this should now rather be improved as follows:
| ob_start( | |
| static function ( string $output ): string { | |
| /** | |
| * Filters the template output buffer prior to sending to the client. | |
| * | |
| * @param string $output Output buffer. | |
| * @return string Filtered output buffer. | |
| */ | |
| return (string) apply_filters( 'gutenberg_template_output_buffer', $output ); | |
| } | |
| ); | |
| ob_start( | |
| static function ( string $output ): string { | |
| // When the output is being cleaned (e.g. pending template is replaced with error page), do not send it through the filter. | |
| if ( ( $phase & PHP_OUTPUT_HANDLER_CLEAN ) !== 0 ) { | |
| return $output; | |
| } | |
| /** | |
| * Filters the template output buffer prior to sending to the client. | |
| * | |
| * @param string $output Output buffer. | |
| * @return string Filtered output buffer. | |
| */ | |
| return (string) apply_filters( 'gutenberg_template_output_buffer', $output ); | |
| }, | |
| 0, | |
| PHP_OUTPUT_HANDLER_STDFLAGS ^ PHP_OUTPUT_HANDLER_FLUSHABLE | |
| ); |
What?
This implements a resolution for this task from #60951:
Why?
As discussed, the code is currently injecting an additional
<body>tag after the BODY has already been opened, like so:Surprisingly (to me at least), this actually works:
Maybe I shouldn't be surprised given HTML's loose parsing rules. But I can imagine other plugins that try to parse the HTML to, for example, apply various optimizations would get might confused when encountering multiple BODY tags.
The attributes could rather get injected on the existing BODY tag instead with the HTML Tag Processor. I see this is not currently possible since the BODY tag is hard-coded in
template-canvas.php. This brings up the question again of output buffering for the entire template (Core-43258). It's something I've been working on in the context of the Performance team's Optimization Detective plugin which output buffers the entire template and then uses HTML Tag Processor to do optimizations.How?
This PR copies the same output-buffering approach taken from the Optimization Detective plugin. Since there is no existing filter for the rendered template output, it starts an output buffer immediately before the template begins rendering. This is at the
template_includefilter.Testing Instructions
data-wp-interactiveanddata-wp-contextattributes appear on the BODY. (There should be only one<body>start tag when looking at the HTML source.)