Looks like there was some kind of runtime conflict when running @emotion/css and @emotion/react at the same time in this app? Some styles would just get clobbered, making things look all weird.
Here, I've removed our @emotion/css dependency, and use the `<ClassNames>` utility element from `@emotion/react` instead. I'm not thrilled about the solution, but it seems okay for now...
...one other thing I tried was passing a `css` prop to Chakra elements, which seemed to work, but to clobber the element's own Emotion-based styles. I assumed that the Babel macro wouldn't help us, and wouldn't convert css props to className props for non-HTML elements... but I suppose I'm not sure!
Anyway, I don't love this syntax... but I'm happy for the site to be working again. I wonder if we can find something better.
Oops, creating a new `SpeciesColorPicker` fn on each render meant that React treated it as a whole new dropdown each time. I've extracted it out into a stable component class, and just pass in the extra props now!
This bug caused changes to kick you out of focus for the dropdown, because it had unmounted and remounted.
This query was very slow! I added an index, and now it's fast!
This code change doesn't actually affect anything, but the comment helps explain what happened, since the index isn't stored in code. (Todo: should I start defining some indexes in our setup files?)
We do animation detection during the preload now, but this wasn't always working correctly: some movies don't actually fully mount the children onto the stage until we start playing. This caused the play/pause button to be missing on the outfit page and the item page, but the animations would still play, depending on the user's saved play/pause state in localStorage.
I saw the short-near-the-front and it just frankly looked awkward? Not sure why I liked it before?
I think this medium at the end of the list is better aesthatically, though it's starting to get a bit messy with the different colors mixed around… but I think there's also a semantic argument that we're keeping the facts about the item together, and the _user-specific_ stuff separate at the end… (putting it at the front would be a good semantic argument too, but I think the NC/NP alignment is too important)
In a previous change, I moved the margin for item badges onto an ItemBadge element… but I didn't think through how that would break the spacing for the loading state of ItemPage. Now, the loading skeleton items _contained_ the badge margin, and so the spacing between badges was shiny skeleton-y.
Here, I replace ZoneBadgesList with a function that just returns the elements, and go back to using Chakra's Wrap component. That will apply the margin to direct children, and the zone badges are direct children now.
One option I'm thinking of in hindsight is an idea I had earlier: Chakra hacks the margin onto _React_ children, but could we use CSS direct child selector instead? A bit trickier to resolve the margin size to the theme's value, but plenty doable… something to consider!
In the previous impl, the buttons variant of the menu would appear on first render, and then the breakpoint stuff would adjust and re-render as the compact nav menu. Now I'm using CSS to show/hide instead!
"Beautiful Green Painting Background" wasn't loading! https://impress-2020.openneo.net/items/75594
```Error building movie clips Error: Expected JS movie library http://images.neopets.com/cp/items/data/000/000/491/491273_31368b3745/491273_2_HTML5%20Canvas.js to contain a constructor named _491273_2_HTML5%20Canvas, but it did not: ssMetadata,Bitmap3,Bitmap5,CachedTexturedBitmap_4183,CachedTexturedBitmap_4184,CachedTexturedBitmap_4185,CachedTexturedBitmap_4186,CachedTexturedBitmap_4187,Symbol20,Symbol8,Symbol4,Symbol7,Symbol2,Symbol1,Symbol9,Symbol2copy,Symbol2_1,_491273_2_HTML5Canvas,properties,Stage```
We already had code to strip out spaces, but not encoded spaces like %20. Now, we decode the URL first, so that space-stripping will work even if it was encoded.
That is, if you're browsing a trade list and you go "oh actually, I _do_ want that!", and click the item page to mark it, then click Back, we'll now update the matching stuff on the trade list page to reflect that it's now a match.
This was just a matter of simplifying the GraphQL query, I think the `currentUserOwnsThis` and `currentUserWantsThis` fields just didn't exist at the time?
We _don't_ yet update your _own_ trade list, if you click through to an item to remove it or something like that. The cache update function isn't too tricky, but it's a bit verbose to implement in Apollo, so I'm not bothering right now!
Oops, the <Wrap> component is nice, but it uses React.Children to apply margin to its _direct_ children, and our badges are not always direct children! (See the new `ZoneBadgeList`.)
I poked my head into how `Wrap` works, and it's honestly pretty simple, so I've applied the same styles manually. Ta da!
Not sure why movie clip building is failing! But it happened outside our try-catch, so it left us in an infinite spinner state.
The repro item is the Spring Topiary Garden Background!
The AMFPHP gateway's json.php endpoint has always had a problem parsing pets whose names start with digits… I've dug into it before, and checked again today, and there really is just no way around it: d584b58e95/core/json/app/Actions.php (L43)
And there aren't any reliable AMFPHP Node libraries out there to make the actual native AMF call.
Buuuut! In today's investigation, I noticed the xmlrpc.php endpoint for the first time. And, wouldn't you know it, there's //great// reliability for something as enterprise-standard as that!
So here, I've switched over to using an xmlrpc client library, which simplifies our calling code //and// makes number pets work correctly 😁 I wouldn't have done it just for the simplification, I think bringing in a library is net more complexity… but getting this finally right is a big relief.
When you navigated directly to ItemPage, the new `safeImageUrl` function would crash during the loading state, because it was trying to safe-ify `undefined`.
Now, I've just made `safeImageUrl` more resilient to that particular kind of unexpected input, by passing through null-y values without change.
When we decided to start out with /api/assetProxy, we didn't know how much the load would be in practice, so we just went ahead and tried it! Turns out, it was too high, and Vercel shut down our deployment 😅
Now, we've off-loaded this to a Fastly CDN proxy, which should run even faster and more efficiently, without adding pressure to Vercel servers and pushing our usage numbers! And I suspect we're gonna stay comfortably in Fastly's free tier :) but we'll see!
(Though, as always, if Neopets can finally upgrade their own stuff to HTTPS, we'll get to tear down this whole proxy altogether!)
I guess something got more picky about the loading sequencing: the fade in animation was happening faster than the cached image could load. Now, we explicitly wait for the image to load (even though we know it's probably cached) before fading it in.
I noticed that, if you're _reading_ the beta callout it's obviously a feedback link, but it's easy to glaze over "Tell us what you think". Here, I've added the word "feedback" to make it stand out on scanning the page, while adding "Got ideas?" to keep it feeling colloquial.
Oops, our movie layer promises don't have a .cancel() method, so calling it crashed our error handler. Now, when there's an error loading a layer and there are HTML5 layers visible, we'll correctly show the "Could not load preview. Try again?" message.
So I broke the Download button when we switched to impress-2020.openneo.net, and I forgot to update the Amazon S3 config.
But in addition to that, I'm making some code changes here, to make downloads faster: we now use exactly the same URL and crossOrigin configuration between the <img> tag on the page, and the image that the Download button requests, which ensures that it can use the cached copy instead of loading new stuff. (There were two main cases: 1. it always loaded the PNGs instead of the SVG, which doesn't matter for quality if we're rendering a 600x600 bitmap anyway, but is good caching, and 2. send `crossOrigin` on the <img> tag, which isn't necessary there, but is necessary for Download, and having them match means we can use the cached copy.)
Oops, I shipped with the images.neopets.com TODO undone! Also, the white background was intersecting with the close X for the feedback form.
In this change, we move the xwee image into our bundle instead of depending on images.neopets.com, and we edit it to have a transparent background, which looks nicer for dark mode. (And we do a srcset!)
Previously I tried to be clever and pre-optimize by putting all the layers onto one canvas… I think this probably helped by batching their paints, but it made fades less smooth by not taking advantage of native CSS transitions, and it made us dip into JS way more often than necessary.
Here, I take the simpler approach: just layers of <img> and <canvas> tags, with each animated layer on its own canvas, and letting the browser handle transitions and compositing, and separate `setInterval` timers to manage their framerates.
I have a suspicion that batching the paints could help performance more, but honestly, maybe that batching is already happening somehow, because things look pretty great on my big-screen stress test now; and so if it _is_ relevant, I want to wait and see after testing on low-power devices.
Not 100% sure on the copy, but I like that it's a bit clearer about the value prop. I tried to work in customization for SEO, but it feels too clunky in a sentence, might need to put it elsewhere in the copy!
Yeah, mm, turns out I don't think it's actually viable to model from Impress 2020, because we can't reasonably set up the SWFs and PNGs in the ways we need, especially for compatibility with Classic DTI.
We can turn this on again later, once Classic DTI is gone, and all assets are converted to HTML5 -- or if we build some kind of bridge to Classic's asset code, or we write new PNG conversion code.
These changes are most relevant for playing around in the dev server, modeing against an empty database. But they'll also help in real-world modeling scenarios! e.g. modeling a new species/color combo is now a bit nicer, we don't show a blank entry in the color picker
this is the last one to get parity with current modeling, I think?? I'm gonna add one more feature though: removing no-longer-used assets from the item
Oops, when building the Support tool to label pet appearances, I didn't realize that there's also a boolean `labeled` field that needs to be true for labeled appearances. Without it, the old app shows the appearance as "Unlabeled".
I also ran this query to fix the rows we'd incorrectly written:
```
mysql> UPDATE pet_states SET labeled = 1 WHERE mood_id IS NOT NULL;
Query OK, 158 rows affected (0.14 sec)
Rows matched: 19640 Changed: 158 Warnings: 0
```
This is mostly because I want to chain the rels after both items and assets save, and I want to be able to specify that stuff a bit more precisely, rather than the like, layers-of-awaits we were building up.
yeah, I had unified Pet into Outfit, but now I think that was overly clever… 😅
Here, I define a new Pet type, and it has some of the fields of Outfit and the deprecated fields still.
I did this because I want petAppearance to work, for UC testing!
We download the schema from prod, and omit real data, but I didn't notice that we were still pulling the metadata of the auto increment counter for IDs! Now, we scrub that from the schema file we save.
I think it's great that we hide the button when it's not relevant, but that makes it hard to know that it exists. Here, we do some cute tricks to blink up the "Paused" button when it first appears, even if the user doesn't have the controls visible right now
We did this a while back too, but I guess something changed in Apollo: I guess it used to return identical item objects from the cache on its own, and now it returns brand new item objects. So we gotta do the object caching hacks ourselves!
This speeds up add/remove item state updates from 500ms to 100ms on my Mac, because we stop re-rendering all the Item components and their complex Chakra children.
This is especially worth doing now, because animations make long updates much more noticeable! (It interrupts the animation 😅)
Honestly kinda surprised this worked on the first go! I was worried something about the process would make the sorta like, instant-cache expectation not work.
Still thinking it might be considerate to like, keep a LRU cache of MovieClip options, so that we don't double-execute these scripts when adding stuff… we even re-execute the ones already applied lol 😅 and that adds lots of script tags to the body!
But yeah I'm not gonna push on it yet until I see evidence that it actually causes performance issues in practice
This is really very cute, but too many items it turns out are lod despite not actually being animated 🙃 it's helpful for looking for test cases tho, so I'm keeping it, but support only!
I also ended up really liking the icon-badge+tooltip design as a way to summarize lil things, so I'm trying Own/Want short badges in the same style.
reflecting further on the abstraction, I'm noticing that this isn't an Easel abstraction like I envisioned early, and that we're baking some Neopets stuff into it. And I think that's the right call, esp with the tricky MovieClip stuff coming up, where I think more barriers would hurt more than they help. So, a new name!
Did it revert? Or did I just never notice that it only worked on mount, not on new loading states?
Also, fixed a bug where we were injecting the script tag way too much, and triggering loading too much that way too!
still just for static stuff, but it's good to be working!
PosePicker got a bit broken, CSS scaling doesn't work quite right anymore, we might need to just up the internal resolution or something?
notable layout change is that the text content will now try to center itself, and we push the buttons off to the right. we also needed to tweak the layout code a bit to get the buttons to feel centered with the top two lines, bc centering against the full block just feels wrong, they want to be top-y in terms of positioning, but still feel centered-y in terms of visual balance
Honestly I don't think this is the long-term home for the Modeling link, I think it'll become a homepage-only link as we add more modules there. But I wanted to get it out of the way!
This wasn't actually super helpful to read anyway, and I think it was causing us to hit rate limits.
We can maybe add back a limited version to like, add path context of _where_ a span happened in the GQL tree, but like, I feel like that's typically been pretty intuitive so far.
Boom, now we can also run a clean MySQL test db on each test that wants it :)
the test I wrote as a sample is currently marked `it.skip` because it's not passing yet!
This updates the MySQL procedure to get the important special colors, but keeps the GQL behavior the same by only filtering to Blue. Just an incremental step before changing the behavior, to make sure I've gotten it right so far!
Snapshots significantly updated, but, from scanning it, I think that's expected changes from actual modeling progress. Hooray!
Fading the whole preview to a black overlay for the loading state was feeling aggressive, especially since loading delay wasn't working correctly!
In this change, I fix loading delay, and I add a nice subtle "corner" variant for outfit preview spinners :)
Here, we add loading skeletons to lots of individual elements, instead of doing a whole item placeholder skeleton. That helps when coming from pages where we have some data, like name and thumbnail, but things is isNc are still missing.
It looks nice, but also particularly means we can handle the loading for the preview separately, get that started faster and iterate better on it in dev!
Initially the spinner was only used in OutfitPreview, where the background was always pretty dark. Now that we use it in more general contexts, we need a light/dark distinction!
Also went and standardized out the `size` props
Here, we extract a lean WardrobePageLayout component, so that we can bundle it into the main app as a loading state for WardrobePage.
This means that clicking Start from the homepage will, instead of flashing the screen to white while WardrobePage loads, show the correctly-sized black/white page layout instead.
I think the Chakra upgrades made these overrides stop working? added !important so that they happen again!
The regression meant the homepage looked worse, always having the selects fade in :/
I'm using my first ever MySQL Store Procedure for clever cleverness in caching the modeling query!
I realized that checking for the latest contribution timestamp is a pretty reliable way of deciding when modeling data was last updated at all. If that timestamp hasn't changed, we can reuse the results!
I figured that, because query roundtrips are a bottleneck in this environment, I didn't want to make that query separately. So, I built a MySQL procedure to do the check on the database side!
I think I got all up in my head about direct queries for this one, because of a previous implementation I had in mind, and I forgot that I could just query species and color from the cache by reference without breaking out of the API provided to the cache function!
I also learned in here that I _can_ look up things from the root by doing `readField("allSpecies", {__ref: "ROOT_QUERY"})`, which I struggled to figure out my previous time. I couldn't figure out how to read an uncached field with arguments (I couldn't quite figure out how to build a proper FieldNode, and passing the string form seemed to provide `null` to the `species` cache field reader), but it's probably doable!
(the permissioning happens on the backend in the prev change! but yeah we send the auth token in the headers, so the backend knows who you are and whether to show you private data)
(also it is just owned items not in any list!)
Essentially I want the center-y visual balance around the image, and the name and _one_ line of tags. If there are extra tags, I want that to go down on its own, rather than bringing down the image+buttons to center against them.
The single-line-of-tags case is the most common, and I think it makes things feel more consistent for all the items to stick to that basic layout, while trying to keep that layout feeling center-y
I was getting annoyed by how, when you're using search, trying on an item will remove conflicting stuff, and then if you decide you don't like what you tried the old stuff _doesn't come back_
As of this change, it does! When you start a new search, we save the outfit state, and then whenever you change the items we ask "hey can these old ones safely be re-worn again?" and re-wear them if so.
On the search panel, not all items have the remove button, and it's confusing to have the other buttons be in inconsistent positions!
Move the Remove button to the left side of the list, so that everything else is positioned the same regardless
Okay, we handle the new pages correctly! Still some weird bugs when you send requests near each other? Probably wise to migrate to Apollo's new way of doing this
This is just an implementation thing, but I realized we can just insert the Zone data into the initial Apollo cache, instead of doing weird field definitions
I _do_ still want the @client tags in the queries though, to tell them not to make server requests at all
Been bothered by this for a while!
My hope is that this isn't a notable marginal performance hit—we were already walking the table and doing string ops anyway, I can't imagine adding to that is actually that much of a marginal lift, when the main bottleneck was probably reads. And the perf should be identical for simple single-word queries anyway. But we'll see how it feels!
This was a subtle little thing for a while! If you switch species/color, such that an item doesn't fit the pet anymore, we used to just hide it. Now, we show it in a list, so that you can understand what went wrong, and have the option to remove it.
Previously, if you switched species/color such that one of your items was no longer compatible, we _would_ still apply its zone restrictions to the visible layer set.
In this change, we fix that server-side, since I think it makes the most sense for an empty appearance to be truly empty!
This was bothering me, I'm surprised and pleased by how easy it seems it was to fix? :)
The strategy is just, look for groups that are provably redundant, and filter them out.
I hope it's correct! It's definitely cozier. Kyrii Mage items are good tests, they have a lot of interesting zones!
I figured that we'd want simpler UI in the ItemsPanel when possible… but now that we've got it pretty simple and comfy, I think the consistency is better
These are nice! :)
The `hideSimpleZones` option I'm not sure about yet, but I figure that:
1. For a new user just doing simple outfits, I feel like the double data on the items page just looks silly, so I want to streamline for that
2. But I _do_ want to let the user think about zone complexity when things _are_ multi-zone.
I did also consider just hiding the zone badge for the header you're under, but I figured the consistency of having the item and its badges look the same in all the places in the list was more important.
This is in preparation for hiding bio zone restrictions but showing item zone restrictions!
I also refactor the build-cached-data script substantially, to run GraphQL against the server instead of a custom query.
okay so the PetAppearance restrictions are stored on the asset, because that's how they're defined on Neopets.com too
but I think that's a confusing API, so here I define `PetAppearance.restrictedZones`, which just maps over the layers and aggregates the zones server-side, same as we would have done on the client
I think that's much easier to understand than having layer contain a field, but having to know that item restrictions _don't_ work that way, you know?
Huh, maybe this is a Firefox bug or something? but the container wasn't applying partial opacity correctly to its children, it was only doing 0 or 1, I think maybe because the children weren't static? I refactor here to make the DOM structure a bit more natural, and fade ins work again 🤷♀️
I'm not sure when this regressed, but changing the outfit was clearing out the whole preview and showing an empty loading state, instead of the intended behavior of showing a loading spinner over the old preview. This affected both the home page and the wardrobe page.
Yeah, so, huh. Fixed! I hope I didn't goof anything else with these effect trigger changes though 😅
The reason I'm doing this now is not just that it's annoying, but as a pair with the color change fix from just now, I want those color changes feeling buttery smooth!
Previously, when changing a pet's color, we would refresh the items panel and send a new network request for the item appearances, even though they're all the same. This is because item appearance data is queried by species/color, for ease of specification.
But! Item appearances are //cached// by body ID. So, if this is a standard color, it's not hard to look in the cache for the standard color's body ID!
Now, most color changes are faster and don't flicker the item panel anymore. We do still refresh the panel and send the requests for color changes that _do_ matter though, like standard <-> mutant!
ahh, in a recent change I made glitched states valid for canonical poses, but didn't make the corresponding change here! This meant that I think the PosePicker showed them, but other ways of getting to them didn't work, including the Candy Acara (who is 100% marked glitched) was no longer pickable at all
Oops, opening the dropdown would auto-focus the left arrow button, because it's now the first focusable element! Make explicit that we want the dropdown instead.
I noticed in prod that the Vercel edge cache can show old data in the Support tool right after you edit it and reload the page, which is super confusing!
In this change, we stop caching the endpoint we use for Support tools, so that the Support tools always feel real-time and trustworthy. (The standard pose picker might still be cached, so it could be a bit confusing for that to be out of sync, but at least you can toggle into Support mode and see that your changes happened _there_, so you don't panic that they're _gone_.)
Previously, I was filtering out glitched appearances from the canonical ones.
But now, I'm thinking it's better to serve glitched ones than no data for a pose at all.
I'm inspired by the case of the Candy Acara, which has _only_ glitched appearances in our db, and I'd like to mark them for reference—but then the site would treat it as no data at all.
Easier to move between appearances quickly! I'm adding this in anticipation of a use case of rapidly labeling Unknown appearances—I want it to be easy to label one and go to the next!
For most users, I want to hide the pose picker if there's not actually anything to pick from.
But I want Support users to be able to open it and use Support mode, to inspect and label Unknown poses!
In this change, I add new UI to show a question mark pose picker button, and a note explaining the empty picker; but only show them for Support users.
I also made a new `useSupport` hook, to replace `useSupportSecret`, now that I have a use case for "is support user?" that doesn't require the secret. Will migrate the rest!
Previously, toggling between the two PosePicker modes could dramatically disrupt the layout, because Popover didn't know to re-measure itself.
In this change, we add a hacky workaround to simulate a window resize event in moments when we know the content is changing. (The realistic ideal solution would still have these manual triggers, but would use an official API in Chakra to notify the Popper instance directly.)
Still just read-only stuff, but now you can look at all the different poses we have for a species/color!
Soon I'll make the pose/glitched stuff editable :3
Some sizable refactors here to add the ability to specify appearance ID as well as pose… most of the app still doesn't use it, it's mostly just lil extra logic to make it win if it's available!
(The rationale for making it an override, rather than always tracking appearance ID, is that it gets really inconvenient in practice to //wait// on looking up the appearance ID in order to start loading various queries. Species/color/pose is a more intuitive key, and works better and faster when the canonical appearance is what you want!)
similar to the layer zoning tools I just rolled out!
not thrilled about the outfit state hacks here bc of how we cache restrict on the appearance rather than the item, but oh well! this escape hatch is pretty easy and solid, and it's a cleanup for another day
Also did a code split here, now that this file is getting larger, to only load this for support users. I don't actually care about restricting console stuff to support users (I'd honestly rather not), but saving the bytes is worth it I think, since support mode is pretty easy to enter when we need to
This is in support of a caching issue in a hack tool coming next! Without this, the change to ItemAppearance restricted zones would make other ItemAppearance fields go missing (bc our hack tool didn't also specify them), so the query would re-execute over the network to find the missing fields we overwrote with nothingness—which would undo the local hack change.
A dev tool to preview what an item would look like in a different zone!
To use, open the browser console, and type `DTIHackLayerZone(layerId, zoneIdOrName)`, but replace `layerId` with the layer's DTI ID, and `zoneIdOrName` with either the zone ID or its name in quotes.
Previously, we were using a custom-y `id` field to help Apollo cross-reference `petAppearance` queries with the results from bulk `petAppearances` queries. Now, instead, we deprecate `petStateId`, and start using `id` to have the same stable value!
This is in anticipation of pet appearance support tools: a stable ID will make it easier to edit them, esp changing their pose (which would otherwise have changed the ID!)
Previously, we would load all `petAppearances` in `PosePicker`, and use cache keys to instantly find it again as a single `petAppearance` in `OutfitPreview` after switching poses.
In this change, we instead have `PosePicker` explicitly load all 6 poses as separate `petAppearance` queries. This simplifies cache sharing between the two components' queries: Apollo can do it automatically, because they were queried the same way in the first place.
I'm doing this in preparation for changing the `id` field of `PetAppearance`, to become `petStateId`. This will help me build pet appearance support tools, by giving the appearances stable identifiers that won't be affected by editing which pose an appearance is!
Huh, some 8-bit species are broken and use the standard body ID!
This was causing our body name query to prioritize 8-bit for standard assets, as the alphabetically-first compatible color; but 8-bit isn't marked standard, so the function kept it labeled 8-bit.
This should fix it and show "Standard Draik" when deleting an asset off the standard draik body!
In practice I saw that this doesn't actually tell you what you _really_ want to know about where the change happened! You want to know it was broken on the Acara or w/e.
Dice reported this, thank you!
My mistake here was that `loadImage` _does_ reject when the image fails to load… but it ends up throwing `undefined`, since I forgot to pass the error along from `onerror` to `reject`!
So we would cancel stuff, but then store `undefined` as our error in state, which our component interprets as no-error.
I tested this by using Firefox DevTools request blocking!
On the homepage, I want to keep the ability to enter invalid species/color pairs, so that you can say "Alien, Aisha" instead of having to pick the Aisha first.
But on the wardrobe page, we were rejecting invalid state changes anyway, so I decided to remove invalid color options from the list. And I added an ability to still switch to any species, and potentially resetting to a basic color automatically to match.
Oops, our loading state logic was eating the error case! I'm not sure exactly where the gap was happening, but I've rewritten the states to be a bit more foolproof, since that first condition was confusing I think.
Ahaha I fucked up a bit! I was indexing into the array of cached zones, instead of looking up by ID. This meant that all zone names were wrong, and some search results weren't loading bc there was no zone data!
I made a fix here, and also added some fallback values, so that if there's an issue in the future we can at least fall back more gracefully than the infinite-spinner case we had here.
In this change, we cache the zones table as part of the JS build process. This keeps the database as our source of truth, while aggressively caching the data at deploy time.
See the new README for some rationale!
I tested this by pulling up dev Honeycomb, and observing that we no longer run db queries to `zones` in the new traces for the wardrobe page. (It's a good thing we did it this way, because I noticed some code in the server that was still loading the zone anyway, and fixed it here!)
This reverts commit 0f7ab9d10e.
The Production Vercel deploys don't seem to like how I did this build trick, even though the Preview deploys seem fine with it 🤔 Reverting for now, sent a message to Vercel support.
Here's just some simple caching: we try to load the asset manifest from the db with the rest of the asset. If it's not present, we load it via HTTP, and write it to the database.
I might try to do a bulk write of manifests at some point, too.
This is because I noticed that one of the main bottlenecks in most of the endpoints now (and definitely the highest-variance) was loading from images.neopets.com.
Another approach I considered was HTTP/2 to load the manifests, because it kinda looks like the server is refusing to open all these sockets at once and effectively does the requests in waves? But images.neopets.com doesn't support HTTP/2 right now anyway, so oh well! (And that would have probably cut us down to ~250ms of HTTP time still, instead of ~600–700. Also, why is network out of Vercel so slow? :p)
I noticed that, while looking up zone data from the db is near instant when you're on the same box, it's like 300ms here!
In this change, we start downloading zone data into the build process. That way, we can have a very fast and practically-up-to-date cache (I'm not sure I've changed it in many years), while being confident that it's in sync with the database source of truth (for things like join queries).
