Drime and code statistics

New features and improvements since 0.5.2, including the changes in 0.5.3 and 0.5.4.

Improvements and bug fixes

• Faster Drime uploads. New-file copies register the uploaded object directly under its final name, avoiding the staging-entry rename delay. Uploads also send a supported file hash with a compatibility fallback.
• Faster folder uploads to Drime and S3. File writes can create their parent folders as they publish, reducing separate directory-creation requests. Empty folders are preserved, concurrent Drime uploads coordinate parent creation, and unsupported cases fall back to explicit folder creation.
• More reliable Code statistics. Fixed line-counting edge cases, preserved completed LOC results when complexity analysis cannot finish, and bounded background work to the views requesting it. Reports retain sorting, selection and scroll position during updates, distinguish partial results from complete reports, and no longer crash when resized.
• Correct audio and model routing. AIFF files opened from archives or remote locations retain the right decoder. ZIP-based model checkpoints open as model previews, and source files such as go.mod retain their text preview.

New features

• Code statistics reports code, comment, blank and physical lines across a folder's whole subtree. Open it from Menu → Code statistics… or a folder's context menu. Separate Totals, Languages and Files tabs show source totals, markup/config totals and per-file details, with options to include generated, data and git-ignored files. Cognitive Complexity and function counts appear where supported. Tables are sortable, the report updates as files change, and Copy report follows the inclusion options.
• Lines of code throughout the browser. Details has an optional, sortable Lines column; Quick view and file tooltips show line counts; the status bar totals selected source LOC. Filter with expressions such as loc:>500, loc:10..100 and -loc:>500. Bubbles and Treemap offer Area by LOC and Code complexity per kLOC colouring. The --complexity command and cadri_code_stats MCP tool also report LOC.
• Torrent content verification. The Torrent viewer finds the files beside a torrent, in a chosen content folder or in the other pane's folder, and shows which are present. Check verifies v1 piece hashes and v2 Merkle roots, with per-file percentages and a piece strip. Missing neighbouring data is reported as not checkable. The file table supports sorting and filtering, and the info hash appears in the viewer and Quick view.
• Tracker scraping on demand. The Torrent viewer's Scrape trackers button retrieves seed, peer and completed-download counts from supported HTTP and UDP trackers after confirmation. Opening a torrent and checking local content contact nothing; scraping sends the info hash and never announces or downloads content. Torrent metadata also recognises legacy text encodings, web seeds and DHT nodes.
• Model-weight and tensor previews. Quick view shows header metadata for safetensors, GGUF/GGML, NumPy, PyTorch checkpoints, ONNX, TensorFlow, TensorFlow Lite, Core ML, HDF5, Keras, XGBoost and pickle containers. Depending on the format, the card shows tensors, parameter counts, data types, quantisation, architecture, producer, inputs and outputs. Reads are bounded and use headers and metadata only; models are never run and pickles are never executed.
• More audio previews. AIFF/AIFF-C, AC-3, MP2 and AMR join the waveform and playback viewer where decoding is supported. MIDI, tracker modules, Monkey's Audio, WavPack, True Audio, Musepack, TAK and DSD containers get header information cards that explain why a waveform is unavailable.
• Playlist tables. M3U/M3U8, PLS, XSPF and WPL open in the Grid viewer with titles, locations, declared durations and availability. Relative paths resolve from the playlist's folder. Local entries can be checked for presence; remote entries are displayed without fetching them.