Serge Huber created UNOMI-1023:
----------------------------------
Summary: Websites that load the Unomi tracker file directly get no
tracking at all
Key: UNOMI-1023
URL: https://issues.apache.org/jira/browse/UNOMI-1023
Project: Apache Unomi
Issue Type: Bug
Components: unomi-tracker
Affects Versions: unomi-tracker-1.5.0
Reporter: Serge Huber
Fix For: unomi-tracker-1.6.0, unomi-tracker-2.0.0
h3. Summary
A website that adds the Unomi tracker to its pages by loading the tracker file
directly gets no tracking at all: no visitor data reaches Unomi and no
personalisation is shown. Nothing visible tells the site owner that it is not
working. The tracker bundled with the Unomi server is not affected, but sites
that use the published tracker file are.
Technical detail: the UMD build (dist/apache-unomi-tracker.umd.js, the package
"browser" entry) throws "Buffer is not defined" in a browser, so the tracker
never starts. The crawler detection library es6-crawler-detect calls
Buffer.byteLength in isCrawler(), which startTracker() always runs. Buffer is a
Node API that browsers do not have. This is also true of the published 1.5.0,
and of es6-crawler-detect 4.x. The tracker bundled with Unomi
(extensions/web-tracker) works only because its build adds Node polyfills.
Bundlers that add no Node polyfills (Vite, webpack 5) probably fail too.
The goal of this ticket is that the tracker builds run in a browser without any
polyfill, without changing anything for the sites that work today.
h3. Decided approach
* When Buffer exists (Node, or a bundler polyfill such as the one in Unomi's
bundled tracker), es6-crawler-detect is called exactly as today, so those sites
see no change.
* When it does not exist, the library gets a Buffer.byteLength, with the same
UTF-8 byte count as Node's, for the duration of the call only. The call is
synchronous, so no other code on the page can see it, and nothing is left on
window (some libraries change behaviour when they find a Buffer).
* No upgrade of es6-crawler-detect on the 1.x line, so crawler detection stays
exactly the same there.
* The tracker sends the same requests as before.
h3. How to test
* Unit tests: the byte count matches Buffer.byteLength on edge-case strings
(emoji, unpaired surrogates, the 4096-byte limit); crawler detection gives the
same result with and without Buffer, and leaves no Buffer behind.
* yarn test:servers: the test page loads the tracker with no polyfill, against
every supported server.
* yarn test:servers compare-1.5.0: the requests are identical to those of the
published 1.5.0.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
