Pavatar 0.3.0 explained (Candidate Recommendation, 2006)
Version 0.3.0 is the last and most complete version of the Pavatar protocol: a Candidate Recommendation dated 18 December 2006. It defined who provides and who shows an avatar, what a valid image is, how a blog finds it and how long a copy may be kept.
This is my explanation of version 0.3.0, section by section, with a few short quotations. The original text is © 2006 Jeena Paradies, editor of the specification; read it in full in the Wayback Machine copy.
Status: stable enough to build on
0.3.0 was the first version that asked developers to implement it. The status section said the Candidate Recommendation was published “to indicate that the document is believed to be stable and to encourage implementation by the developer community”. Comments went to a project address on pavatar.com instead of Jeena's personal one, which is a small sign that Pavatar had become a project of its own. The earlier drafts — 0.2.0 and 0.1.0–0.1.2 on jeenaparadies.net — are listed as previous versions.
Who is who
The specification names four parties:
- Provider — the person who leaves a comment and owns the avatar.
- Provider's website — the HTTP address the provider gives as their website.
- Consumer — the site that shows the comment and the avatar.
- Implementation — the software on the consumer's side that does the work.
The provider and the consumer may be the same person: a blogger answering comments on their own blog is both.
What counts as a valid image
The technical definition is short. A valid Personal Avatar is an 80×80 pixel GIF, PNG or JPEG, publicly reachable through a valid HTTP URL and served with the correct Content-Type. The size limit came with its own arithmetic:
A valid Personal Avatar must not exceed the size of 409600 Bit (the octuple size of an uncompressed 80x80pixel big picture with 8-bit color depth).
Personal Avatar 0.3.0, section 2.a. © 2006 Jeena Paradies.
In bytes that is 51,200 — far more than an 80×80 icon needs, which made the limit easy to meet. A provider who did not want the image used could refuse with HTTP 403, and the recommended way to refuse delivery completely was the keyword none in the X-Pavatar header.
Autodiscovery, step by step
Section 3 lists the three ways to publish the image address and section 4.a turns them into an algorithm. Given the address from the comment form, the consumer:
- Reads the HTTP headers of the response. If there is an
X-Pavatarheader, its value is the avatar. Clients that can read headers must do so, because headers override link elements when the two differ. - Searches the HTML for the first match of
<link rel="pavatar" href="([^"]+)" ?/?>and decodes the four allowed entities (&,<,>,"). - Looks for
pavatar.pngin the directory of the provider's page and, if it is not there, at the document root.
If any step finds none instead of a URL, the lookup is aborted — the specification is explicit that there will be no avatar at all. The diagram on Pavatar explained replays a lookup through all three steps, and the Pavatar check page gives the same steps as curl commands.
The informative appendix suggests how to keep this cheap: start with a HEAD request in case the header is enough, stop reading the HTML at </head> or <body>, or fetch only the first few kilobytes, because the link element belongs near the top of the document.
After discovery: download, check, keep a copy
What a blog did with the image
example · spec 0.3.0- Avatar URL · the result of discovery: the header, the link element or pavatar.png
- Download · the blog fetches the image once; serving it from the commenter's site on every page view is discouraged
- Check · 80×80 pixels, GIF, PNG or JPEG, at most 409,600 bits, the right Content-Type
- Cache · a local copy, refreshed from Cache-Control or Expires, or weekly when the site sends neither
- Show · the cached copy beside the comment; resizing is allowed, changing what the picture shows is not
0.3.0 asks consumers to use as little traffic as possible. In practice that meant:
- Cache everything you serve. The implementation should keep its own copy and should not serve the avatar from the provider's site.
- Refresh by the rules. Check for a changed image according to the
Cache-ControlorExpiresheaders; if the site sends neither, check weekly. Conditional requests such asIf-Modified-Sincekeep the refresh cheap. - Remember the misses. A website where discovery found nothing should not be checked again until the provider posts something new.
- Resize, but do not rewrite. The consumer may scale the picture or change its colors, but
it must respect the visual identity of the provider and therefore avoid destructive transformations of the Personal Avatar
Personal Avatar 0.3.0, section 4.b. © 2006 Jeena Paradies.
The appendix also allows an automatic “denial-of-publish” for avatars the system does not know, optionally with a notification — a moderation hook rather than a rule.
What changed from 0.2.0
| Topic | 0.2.0 | 0.3.0 |
|---|---|---|
| Roles | commentator, publisher, webmaster | provider, consumer |
| Status | early, not for implementation | Candidate Recommendation, for implementation |
| Size limit | 102,400 bits | 409,600 bits |
| Caching | the commentator could forbid it with Expires or Cache-Control; the consumer could also link to the original | the consumer should always cache and should not serve the image remotely |
| Resizing | allowed unless caching was forbidden | allowed, but must respect the provider's visual identity |
| Misses | not covered | remember sites without an avatar until the next post |
| Contact | Jeena's personal address | a project address on pavatar.com |
The discovery order — header, then link, then pavatar.png — and the 80×80 format did not change.
Who implemented it
The Perl module Net::Pavatar on MetaCPAN still links to this version's address, which is one reason this page lives at the same URL. Other implementations and the reactions to the protocol are listed on Pavatar explained.
Back to Pavatar explained or the Pavatar archive.