Proposed additions that have not landed yet. API names and examples are
proposals, not existing interfaces or a finished specification, and there are
no timeline commitments. What has landed is documented with its module: the
cli, io and process modules, fiber-aware stream and process waits, and
standalone bundles (tools/bundle.mob).
| Priority | Addition | Intended home |
|---|---|---|
| Next | File and directory watching | Existing os module |
| Next | CSV reading and writing | Bundled csv module |
| Later | Bundles that read their scripts and assets in place | Bundle launcher |
| Deferred | HTTPS, WSS and TLS transport | Existing networking modules; backend undecided |
| Low priority | Typed arrays and native data operations | Optional data package |
Keep the core interpreter small, use the existing module and package system, and design native functionality for Linux, macOS and Windows.
Extend os with file and directory watching, filtering and debouncing.
Deliver events through a channel, so scripts can rebuild templates, regenerate
code or reload assets.
Define recursive watching, rename events, atomic file replacement, overflow
notification and watcher shutdown. Normalise useful behaviour across the OS
backends (inotify, FSEvents, ReadDirectoryChangesW) without promising identical
low-level event sequences. Waiting for events should suspend only the waiting
fiber, as io and process waits already do.
Ship CSV as a general-purpose bundled module, independent of the optional
data package. Support incremental reading and writing through io streams,
quoting, embedded delimiters and newlines, configurable delimiters, headers,
and record and field locations in errors.
Keep fields as strings by default, with explicit conversion options. Define
how duplicate headers and rows of the wrong length are handled. A future data
adapter can read CSV in typed batches without making CSV depend on it.
A bundle extracts itself to a per-bundle cache directory and runs from there. Reading scripts and assets directly from the embedded archive would remove the extraction step and the cache. Native plugins still need to be on disk to be loaded, so this applies to scripts and assets.
An open architectural decision. The existing HTTP and WebSocket code already handles the protocols; adding HTTPS doesn't require replacing it.
The alternatives considered were a shared TLS stream under the existing
protocols, or libcurl-backed HTTP client requests. Libcurl would not provide
TLS listeners for the web server. If it were chosen, link it statically into
the native module; Windows could use Schannel, while Linux would need a chosen
TLS backend. Avoid adding distributed shared libraries.
Settle certificate trust, hostname verification, client and server integration, shutdown and timeouts when this work resumes. Neither libcurl nor OpenSSL is a settled dependency.
Specify this separately. The direction is an optional package with native bulk operations over typed buffers and a convenient Mobius API. OpenBLAS (numerical operations) and nanoarrow (columnar data interchange) are candidates, not chosen dependencies.
Keep buffer ownership, view lifetimes and resizing rules explicit. Avoid copying where layouts allow it, without promising zero-copy everywhere. CSV stays useful on its own in the bundled library.