go-libwebp v2.3.1 depended on its oen v1.8.0; v2.3.2 happily fixes this!

Jack very kindly agreed to fix the versioning issue with the libwebp library, which means v2 is now available to us!

https://git.sr.ht/~jackmordaunt/go-libwebp/commit/584e42fc5f8fe3650d01cd1ce3e73acc27d5f91d

There's an occasional bug with the JPEGs within SVGs, in that they don't decode properly. I need to write tests and investigate.

I still haven't figured out what's constantly flipping these in my IDE, and for this project only. I'll try and investigate.

Swaps to a newer tiff library, which supports macOS scanner TIFF images 🎉

Makes the back of the postcard just the card colour, and uses the shape of the front of the postcard (transposed by the card thickness amount) for the back.

Setting the version in a version_test file means that version number is set for all tests. I've also swapped the embedded version number in all fixtures, so that they align.

Also switches to using build tags for default postcard format on postoffice. This allows the postoffice server to provide webp images.

Includes: handling postcards with no back, commented out (!) code for not-quite-working bezier curves, ensure ignored transparency images have no transparency enocded

This is a hard decision; the XMP data is slightly larger than the JSON metadata *anyway*, and the JPEG data is expanded by 4/3 in an SVG (Base64 encoding), so this increases the size of the svg format by ~2kb. But it does mean that the jpeg within the SVG is a fully contained postcard in its own right (minus the border — which I may want to store in the metadata too?)

I really want to figure out what's going on here

The WebP library I'm using accepts 0-1 as quality values, but the underlying transpiled libwebp expects 0-100, so my "70" is producing lossless images every time!

I've informed the author, and hopefully a similar fix will be included upstream soon. I've vendored a modified version of this library here until then.

Pre-hidden secrets don't need to be re-hidden.

Because the web codec now returns the HTML and CSS, the USD codec was unsure which of the files provided was the image it should be using as a texture. This is now fixed!

Moving to a subdirectory *combined* with service worker caching has caused some issues — this ensures the trailing slash is consistent, and hopefully will resolve the issues when caches expire. I have more work to do on ensuring service workers work consistently well…

I forgot to add CardColor to the metadata stored in XMP. The recent version number fix means XMP tests are now fragile, they will need fixing.

Was reading the version number from the previous iteration of this package, because of accidental import loops. Made a hacky fix here, but I'd like to fix this so the version number can stay in the root, if possible.

These were attempting to allocate a negative number of bytes, as the padding must be *at least* 4 bytes long to be able to hold the FileHeaderID and extra field length.

Fixes the usdchecker complaint about MaterialBindingAPI not being present. This is still a USDZ that contains an ascii USDA file (rather than a compiled USDC file); but only Apple tools complain about that…

Ensure that the file data is always offset to a multiple of 64 bytes from the zipfile start, as expected by the USDZ spec.

This makes space for a '/convert' endpoint which can also be bundled into the WASM

Given build changes will likely produce chnages to the built assets, they should update the changelog & be built.