Mounting tar archives as a filesystem in WebAssembly

Instead of extracting tar.gz, a JSON index containing the file offsets and sizes within the tar is created and mounted directly into Emscripten's WORKERFS.

This approach leverages tar's 512-byte headers and contiguous data structure. tar-vfs-index reads a tar or tar.gz stream to generate file_packager-compatible metadata, recording the start/end byte range for each file.

The mounting procedure is as follows.

• Fetch the tar.gz and metadata together with fetch().
• Decompress the tar data using the browser's DecompressionStream('gzip') to create a Blob.
• Connect it to the virtual filesystem with FS.mount(WORKERFS, { packages: [{ metadata, blob }] }, '/pkg').

This way, when C/C++ code reads a file, only the needed range of the Blob is sliced, so files don't need to be extracted separately. WebR applied this approach to R package distribution, improving loading speed and memory efficiency while still using plain tar.gz on a static server.

Another approach is to append the index to the end of the tarball using the --append option to distribute it as a single file. The loader first separates the embedded metadata, then mounts it the same way.