Avoid Partial Content HTTP download in webR, fixing browser caching for cairo.so

[georgestagg]

Github Pages supports serving content with HTTP status 206, "Partial Content", but content that is served in this way is not cached by the brower (at least, not by Chrome). So, when using Shinylive hosted on GitHub Pages multiple (slow!) network requests are sent on every page load for the webR resource file cairo.so, even if the file has already been previously downloaded by the browser.

Emscripten is configured to fetch lazy filesystem content with size greater than 1MB in chunks, using the partial content mechanism. The chunk size is hard coded here and I can't see a way to disable things. We might be able to submit a PR to Emscripten implementing such a disable switch, but even so this would then still require upgrading Emscripten for a new release of webR, which we can't do until R 4.5.0 for ABI reasons.

For the moment, this PR patches the lazy file system mechanism to always skip partial content and download the entire file. With this, cairo.so is loaded in a single request, and the response is cached as expected.

R package library content should not be affected, since webR uses a custom VFS mounting routine for those in any case. Also, the patch only touches webR, pyodide should not be affected in any way. AFAICT, the only file large enough in webR's base distribution to be affected in this way will be cairo.so itself.

[wch]

Just curious: how will this affect the shinylive.io site, or other hosting services?

[georgestagg]

It will depend on the hosting service, specifically on if the response to a HTTP HEAD request for content is served with the Accept-Ranges: bytes header set. When set, Emscripten will try to download lazy-loaded content in chunks (the check is here). We're modifying the conditional on hasByteServing so that the code behaves as if Accept-Ranges is never set.

For the shinylive.io site in particular, it does not serve cairo.so with Accept-Ranges right now so there should be no change in behaviour. A quick online search suggests the feature is possible to enable in S3, but is off by default.