Update June 28th: 2.7.6 was released, all is (or should be) fine … :-)
There currently are 2 known issues in Autoptimize 2.7.5 that will be fixed in the next release;
when “inline & defer CSS” and “also aggregate inline CSS” are active then for logged in users the top “admin bar” might become invisible. unticking “also aggregate inline CSS” is a confirmed workaround.
when “inline & defer CSS” is active, CSS-files that are not aggregated (excluded or 3rd party) and that do not have a media-attribute will not be deferred.
As of the soon-to-be-released Autoptimize 2.7.4, all occurrences of “blacklist” and “whitelist” in the code will be changed into “blocklist” and “allowlist”. There is no impact for users of Autoptimize, everything will work as before.
If however you are using Autoptimize’s API, there are two (to my knowledge rarely used) filters that are now deprecated and will be removed at a later stage. `autoptimize_filter_js_whitelist` and `autoptimize_filter_css_whitelist` still work in 2.7.4 but if you’re using them switch to `autoptimize_filter_js_allowlist` and `autoptimize_filter_css_allowlist` to avoid problems when they are removed in the release after 2.7.4.
Small post-publishing clarification dd. 22/07/2020: this post is just an announcement, I feel no urge to discuss the change and am not really interested in arguments pro or contra. Don’t fret over this change, fretting is useless, instead enjoy the summer, kiss your lover, read a good book, … :-)
If you tested your blog’s performance on Google PageSpeed Insights yesterday and do so again today, you might be in for a surprise with a lower score even if not one byte (letter) got changed on your site. The reason: Google updated PageSpeed Insights to Lighthouse 6, which changes the KPI’s (the lab data metrics) that are reported, adds new opportunities and recommendations and changes the way the total score is calculated.
So all starts with the changed KPI’s in the lab metrics really; whereas up until yesterday First Contentful Paint, Speed Index, Time to Interactive, First Meaningful Paint, First CPU Idle and First input delay were measured, the last 3 ones are now not shown any more, having been replaced by:
Largest Contentful Paint marks the point when the page’s main content has likely loaded, this can generally be improved upon by removing removing render-blocking resources (JS/ CSS), optimizing images, …
The total score is calculated based on all 6 metrics, but the weight of the 3 “old” ones (FCP, SI, TTI) is significantly lowered (from 80 to 45%) and the new LCP & TBT account for a whopping 50% of your score (CLS is only 5%).
Lastly some one very interesting opportunity and two recommendations I noticed;
GPSI already listed unused CSS, but now adds unused JS to that list, which will prove to be equally hard to control in WordPress as JS like CSS is added by almost each and every plugin. Obviously if you’re using Autoptimize this will flag the Autoptimized JS, disalbe Autoptimize for the test by adding ?ao_noptimize=1 to the URL to see what original JS is unused.
GPSI now warns about using document.write and about the impact of passive listeners on scrolling performance which can lead to Google complaining about … Google :-)
an experimental option to provide “fallback autoptimized CSS and JS”; in some cases a cached version of the HTML links to autoptimized CSS/ JS that was already removed. if you enable that option, Autoptimize will try to intercept those requests and redirect them to fallback versions to limit the breakage. the option is default off for now, but might become default on at a later stage.
For this purpose Autoptimize hooks into template_redirect and will redirect to fallback Autoptimized CSS/ JS if a request for autoptimized files 404’s.
For cases where 404’s are not handled by WordPress but by Apache, AO adds an ErrorDocument directive in the .htaccess-file redirecting to wp-content/autoptimize_404_handler.php. Users on NGINX or MS IIS or … might have to configure their webserver to redirect to wp-content/autoptimize_404_handler.php themselves though, but those are smart cookies anyway, no?