Early Abort Plus Range-Based Lazy Retrieval

Issue 46 Edition 2026-02-15 5 min read

General

Sources: 1 • Confidence: High • Updated: 2026-02-19 20:51

Key takeaways

Gwtar calls window.stop() early and places inline uncompressed tar content after that call to prevent the browser from downloading the entire file up front.
Gwtar uses a PerformanceObserver configured for resource entries to detect attempted resource loads.
The suggested local workflow is to extract the embedded tar content using a perl pipeline into tar --extract and then open the resulting HTML file in a browser.
Gwtar is a project by Gwern Branwen and Said Achmiz intended to combine many assets into a single archived HTML file that remains convenient to view in a browser.
After stopping the initial load, Gwtar uses HTTP Range requests to fetch required resources on demand from embedded tar data.

Gwtar calls window.stop() early and places inline uncompressed tar content after that call to prevent the browser from downloading the entire file up front.
After stopping the initial load, Gwtar uses HTTP Range requests to fetch required resources on demand from embedded tar data.

Gwtar uses a PerformanceObserver configured for resource entries to detect attempted resource loads.
Gwtar fulfills observed resource requests by locating a previously loaded resource or fetching the needed byte range and injecting it via a blob: URL.

The suggested local workflow is to extract the embedded tar content using a perl pipeline into tar --extract and then open the resulting HTML file in a browser.
Gwtar files cannot be opened locally in a browser due to security restrictions and will display an explanatory message instead of the intended page.

Gwtar is a project by Gwern Branwen and Said Achmiz intended to combine many assets into a single archived HTML file that remains convenient to view in a browser.

What are the exact browser and hosting requirements for Gwtar to work reliably (for example, which browsers support the necessary observer behavior and which servers/CDNs reliably support the needed Range request behavior for such files)?
What is the expected performance profile (time-to-first-render, total bytes transferred over a session, memory growth due to blob materialization) across page sizes and asset counts?
How is the archive produced from an existing multi-asset site (tooling, required transformations, how resource URLs are rewritten, and how tar member offsets are mapped for Range retrieval)?
What security and policy constraints beyond local file access apply (for example, CSP interactions with blob URLs, cross-origin rules, or restrictions that may vary by browser version)?
Is there an intended non-extraction offline workflow (for example, running a minimal local HTTP server) that avoids the local file restriction while preserving the single-file container property?

Growing interest in single file web packaging that still supports lazy asset access could benefit toolchains, hosting, and CDNs optimized for Range requests and large object delivery, if compatibility and performance are acceptable.
If browser APIs can reliably intercept resource loads and remap to blob URLs, more client side resource resolution patterns could emerge, increasing importance of API stability, memory behavior, and security policy handling.

Demonstrated cross browser reliability for early abort behavior, PerformanceObserver based detection, and blob URL injection across modern versions.
Consistent Range request support and correct byte serving across common servers and CDNs for large archived HTML objects.
Benchmarks showing acceptable time to first render, bytes transferred per session, and controlled memory growth across page sizes and asset counts.

Key browsers restrict or change PerformanceObserver resource entry behavior or block required interception and blob injection patterns.
Range requests are unreliable or disabled in common hosting stacks or CDNs for this use case, breaking on demand retrieval.
Security and policy constraints such as file access limits, CSP conflicts, or cross origin rules make non extraction workflows impractical.