We have a frontend that, upon loading, redirects the user to Keycloak. The user logs in and is returned to the application. The code coverage written to the disk was incomplete and after hours of debugging I found the issue. The problem is the caching of references to the window.__coverage__ object. When the code that redirects the user is instrumented, the covered statements will be different for the initial redirect and the subsequent return so this plugin will correctly recognize them as different and will keep a reference to both coverage objects. When the code that redirects the user is not instrumented (because it is a library), the coverage object will be identical. Both coverage objects will have identical statement coverage and the current version of the code will throw away the second coverage object, keeping only the reference to the first one which is not updated anymore. This will result in a code coverage where only the statments until the call to the library are covered.

Before:

----------|---------|----------|---------|---------|-------------------
File      | % Stmts | % Branch | % Funcs | % Lines | Uncovered Line #s 
----------|---------|----------|---------|---------|-------------------
All files |      50 |      100 |       0 |      50 |                   
 utils.js |      50 |      100 |       0 |      50 | 2                 
----------|---------|----------|---------|---------|-------------------
ERROR: Coverage for lines (50%) does not meet global threshold (100%)

After:

----------|---------|----------|---------|---------|-------------------
File      | % Stmts | % Branch | % Funcs | % Lines | Uncovered Line #s 
----------|---------|----------|---------|---------|-------------------
All files |     100 |      100 |     100 |     100 |                   
 utils.js |     100 |      100 |     100 |     100 |                   
----------|---------|----------|---------|---------|-------------------

Keywords for Google: Code coverage different in afterEach.