Changelog

All notable changes to this project will be documented in this file.

[0.1.4] - 2026-10-28

Added

• R2: list-parts returns the parts already uploaded for a multipart upload (following pagination), or nil if the upload no longer exists — lets a server resume an interrupted browser-direct upload.

Fixed

• XML tag matching no longer treats a longer tag as a match for a shorter one (e.g. Part matching <PartNumber>).

[0.1.3] - 2026-09-16

Added

• Cloudflare Images client with local uploads (File, byte[], or InputStream), remote URL ingestion, and browser-direct one-time upload URLs.
• Image management through get, list/list-seq, update, delete, and exists?, plus original-image downloads, account statistics, signing keys, and a raw REST API escape hatch.
• Named and flexible delivery URLs, including HMAC-signed URLs for private images.
• Zero-dependency Cloudflare REST v4 transport and JSON codec.
• Live Images upload/get/delete integration coverage.

[0.1.2] - 2026-06-29

Added

• R2: browser-direct multipart uploads — create-multipart-upload, presigned-upload-part, complete-multipart-upload, and abort-multipart-upload are now public. A server can broker the multipart lifecycle and hand the browser one presigned PUT URL per part, so the bytes of a large object never pass through it.

[0.1.1] - 2026-06-21

First public release. (0.1.0 was burned by a partial Clojars deploy whose pom omitted a license; it is not a usable release — use 0.1.1.)

Added

• Initial project scaffolding (deps.edn, tools.build, test runner).
• R2: bucket handle construction against R2's S3-compatible endpoint.
• R2: put with automatic single-PUT vs. multipart selection — large objects (and unknown-length streams) upload in concurrent, individually-retried parts, aborting the upload on failure to avoid orphaned parts.
• R2: downloads — get-bytes, get (bytes + metadata), and get-stream (streaming, for large objects); all return nil for a missing key.
• R2: head (object metadata) and exists?.
• R2: delete, and delete-many (batched, automatically chunked above the 1000-key S3 limit, reporting per-key errors).
• R2: list (one page, with :prefix/:delimiter/:max-keys and continuation tokens) and list-seq (lazy enumeration across all pages).
• R2: presigned-url for time-limited download (:get) and direct-upload (:put) URLs, via SigV4 query-string signing.
• R2: request escape hatch to the raw signed-S3 surface.
• Internal zero-dependency S3 transport: an AWS Signature V4 signer (internal.sigv4, header and presigned signing, both checked against AWS's published test vectors) over the JDK java.net.http client (internal.s3), with minimal S3 XML handling (internal.xml).

Changed

• Dropped the Cognitect aws-api dependency in favour of the zero-dependency transport above. The library now adds only a ~13 KB jar (plus Clojure) instead of the ~250 KB of transitive AWS SDK artifacts. The R2 escape hatch changed from aws-api op maps (invoke) to raw signed HTTP (request).