You know how it is. You go in a certain direction for a while, perhaps through adversity, only to reach a point and ask yourself, “And exactly why I am doing this, again?”
(Well, if you don’t know how that is: trust me, kids, you’ll reach that point if you live long enough.)
Back in December, I began running this site on a combination of Eleventy and the webpack software bundler. This came after I’d spent a few weeks with the webpack-powered Gatsby and reached the conclusion that I could get eighty to ninety percent of the same performance and flexibility out of an Eleventy/webpack combo while enjoying it—especially Eleventy—’way more than the sometimes maddening Gatsby.
But, then, some funny things happened on my way to SSG bliss.
A pruning process
In the ensuing months, I removed two of the biggest reasons for having added webpack in the first place:
I dropped SASS/SCSS for PostCSS while experimenting briefly with Tailwind CSS. While I didn’t warm up (then) to Tailwind, I liked PostCSS a lot and decided to stick with it. And while you certainly could use webpack to incorporate PostCSS with Eleventy, it wasn’t necessary to do so.
Hero images, once a mainstay of this site, were the next to go after it became clear the hassles of trying to provide the proper sizes and, in particular, file formats weren’t worth the effort when compared to those images’ actual utility. Only those few images necessary to tell certain posts’ stories remained, and their spartan quantities and nature constituted wastes of all the webpack code that processed them—relatively slowly—on each build.1
And I began to wonder just how much I really needed webpack—and whether my site-development process might actually be better without it.
In a similar vein but not because of webpack one way or the other, I was unhappy with the size of the CSS I was delivering with each page; and my background attempts to shrink it through old-fashioned editing and rewriting hadn’t gone well. (I’d also tried more automated approaches but had failed with them, too.)
Still, one intriguing thing I’d learned from the Tailwind experiment was the amazing power of PurgeCSS, when combined with PostCSS, to shrink even the multi-megabyte CSS loads of a Tailwind config down to just a few kilobytes at production. That was three to four times smaller than my “handmade” CSS after each had undergone the same minification process during a build to production.
My more recent rethinking process coincided with a renewal of at least curiosity about Tailwind, which in February became compatible with CSS Grid (this post from November, 2018, tells why that mattered to me) and just a few days ago also began including PurgeCSS right out of the box.
Anyway, I’ll get back to the Tailwind front in a bit.
In the background during all this, I’d noticed that Eleventy itself was getting additional and more powerful plugins and, with newer Eleventy versions both since I’d started using it and those to come in the near future, greater capabilities.
All this gradually made it clear to me that, if I made some reasonably sensible config changes, I could build an Eleventy-based repo that probably could do everything the Eleventy/webpack one could, but without webpack. It would be an interesting experiment, anyway; and, even if I couldn’t make it go the distance, the real site could still exist on the unaltered Eleventy/webpack repo.
With that safe haven in mind, I cloned the existing repo and started the process of unpacking webpack from it. Just as I’d called the existing repo
eleventy_bundler to signify that it was joined at the proverbial hip to webpack, I named this new one
Image processing: birth of a notion
However, I soon ran into what appeared to be an insurmountable roadblock to my ever being able to make
eleventy_solo anything but a locally running experiment: I couldn’t find a workable way to process the few images the site still has.
With the Eleventy/webpack combo, I’d used the responsive-loader webpack add-on to create and then process responsive images, and the lazysizes library to make them “lazy-load” on nearly all browsers. My hope had been to replace them in
eleventy_solo with some Eleventy plugins but, for reasons I still don’t understand, I couldn’t get those plugins to work successfully in the new repo.
So, while I had everything else working pretty well, I was still delivering only one size of each image, and an unprocessed (generally too large) image file at that. And, if I really did ever want to make
eleventy_solo the repo for this site, I wasn’t about to let that be the best I could do.
Then I got to thinking about how
That’s how a script called
imgxfm.js was born. After an embarrassingly high number of false starts, I got it working with the sharp library to produce multiple sizes, and the desired formats, of image files and plop them into
/_site/images/ as I wanted.
imgxfm.js didn’t always produce smaller files, especially where the largest iterations of certain
.png files were concerned, but in the vast majority of cases it accomplished this well enough for my purposes. Most of the site’s larger images are
.jpg files, which it handled just fine, so I was satisfied with the overall result.
The commit in which
imgxfm.js first appeared was called “Image processing while waiting for plugin to be ready”; but, once this little file was reliably doing what I wanted on each build, I realized it was no longer a stand-in. The star of the show had failed to arrive at the performance hall on time (albeit through no fault of his own), so I let the scraggly stand-in take his place; and, scraggly though he was, he did just fine and pleased the crowd just the same.
Update, 2020-05-25 . . .
The next day, I switched to a real image-processing plugin—
eleventy-plugin-local-respimg by Sam Richard—thus replacing
imgxfm.js. However, while this plugin obviously was superior to my script, my later switch from LazySizes to LazyLoad for lazy-loading made it necessary to revert to
imgxfm.js, at least for now. (That said: I still extend a big thanks to Rick G. for helping me get past earlier issues with Mr. Richard’s excellent plugin.)
“An API for your design system”
But how did Tailwind get back into the picture?
Well, one lunchtime, I decided out of curiosity to check up on what’s been going on in the Tailwind universe. While I read articles about it from time to time, I also like to watch video tutorials to see if there is something I’ve missed about this popular framework.
This day, my search brought up a video of a presentation that Tailwind creator Adam Wathan gave at last year’s Laracon US event. The talk was called “Tailwind CSS Best Practice Patterns.” Perhaps the most interesting point Mr. Wathan made, early on, was . . .
“Tailwind is an API for your design system.”
I’d never thought of it that way before. Intrigued by the concept, I decided to give Tailwind another try; so, now,
eleventy_solo would be not only minus webpack but also plus Tailwind.
As a result, even while I was still battling the aforementioned image-processing issues, I borrowed files from old commits dating back to my previous experiment with Tailwind. Of course, that was back when all my templates were in Nunjucks, before my recent changeover to all-.11ty.js templating, so it wasn’t just a matter of drag-and-drop: I spent quite a few hours just rejiggering conditionals and other elements to work in .11ty.js, regardless of the presence or absence of Tailwind.
But, in the end, I felt this retrenching, like the jettisoning of webpack, made a lot of sense. Just as I’d done with the other repos, I planned on making this one public if/when I deemed it mostly ready for prime time, and I suspected that the majority of folks with curiosity in Eleventy (a.) might not want to deal with webpack and (b.) almost surely would have more interest in the increasingly popular Tailwind than in my pedestrian CSS.
A solo performance
Late in the day on May 7, I finally had things ready enough, and “told” Netlify to do a new build—but, this time, from
eleventy_solo rather than
eleventy_bundler as it had been set to do since five months ago. On the first try, a glitch occurred and halted the build. Fortunately, in such a case, Netlify just errors out and leaves your existing site in place. Then, after running my
testbuild script from
package.json locally to figure out what was wrong, I got things to go through a little while later.2
Oh, yes: build time is now routinely well under a minute, while
eleventy_bundler’s build times generally hovered around a minute-forty but sometimes went as high as two-and-a-half minutes. Since this greater speed also extends to my local dev efforts, I am happy to recover those cumulative minutes and hours going forward.3
While the new repo is now public for your viewing and spitballing pleasure, its predecessor also remains in place for anyone who’s still interested in getting Eleventy and webpack to make beautiful music together.4
And that’s fine. I just hope such folks will realize that Eleventy plays a pretty sweet tune all by itself, and its repertoire and orchestra are getting ever bigger and better.
- This wasn’t just a matter of keeping me from twiddling my thumbs every time I made changes to the site. Limiting build times is important—especially since the Netlify free tier has a limit of 300 minutes a month; and, recently, I’ve been using Zapier zaps to auto-build the site each midnight, Central time, to keep its webmentions more frequently updated. The Eleventy/webpack combo typically took roughly two minutes per build, meaning over sixty minutes a month even if I didn’t change anything else on the site—and, to be sure, I am always changing things, even things you may never notice. ↑
- There was one more false start when, after getting the build to work for the first time, my eager inspection of the resulting CSS showed it was a porky 1.6 MB even after the usual minification and GZipping—meaning PurgeCSS hadn’t worked. However, I soon realized that it was because I’d forgotten to set the final build’s environment variable to PRODUCTION rather than DEVELOPMENT (on which I’d received great and patient advice from none other than Mr. Wathan himself). Once that was fixed, the next build went through just fine and the real site’s CSS had dropped to just under 7 KB. Bingo. ↑
- That shorter build time didn’t last long, once I began using a larger number of image sizes for responsiveness. Still, I’ve tried that same number of image sizes with the previous webpack repo and it would clearly be a lot longer than what the current non-webpack repo would do. ↑
- Please understand that, as with the Hugo and Gatsby repos, I’m not promising to keep the previous Eleventy/webpack repo up to date with this one where content and features are concerned. I may in fact do so sometimes for bleeps and giggles; but I’m not committing to it. ↑