project reports `cant find package` for a package that is not referenced at all

project reports cant find package for a package that is not referenced at all

simple reproduction: https://stackblitz.com/edit/typescript-blff19

project as-is works fine, but uncommenting two lines that import and access @tensorflow/tfjs-backend-cpu results in build error:

Can’t find package:@tensorflow/tfjs-node

but @tensorflow/tfjs-backend-cpu does not reference @tensorflow/tfjs-node anywhere
(and yes, @tensorflow/tfjs-node cannot be installed since its a nodejs module while this is a pure browser project)

the only mention of tfjs-node inside tfjs-backend-cpu module anywhere is inside string block:

backend_util.warn('\n============================\n' +
    'Hi there 👋. Looks like you are running TensorFlow.js in ' +
    'Node.js. To speed things up dramatically, install our node ' +
    'backend, which binds to TensorFlow C++, by running ' +
    'npm i @tensorflow/tfjs-node, ' +
    'or npm i @tensorflow/tfjs-node-gpu if you have CUDA. ' +
    'Then call require(\'@tensorflow/tfjs-node\'); (-gpu ' +
    'suffix for CUDA) at the start of your program. ' +
    'Visit https://github.com/tensorflow/tfjs-node for more details.' +
    '\n============================');

is it possible that stackblitz parser misbehaves and does not consider this a string and reports a dependency purely because of this?

also note that using other libraries such as @tensorflow/tfjs-backend-webgl that have a hard dependency on @tensorflow/tfjs-backend-cpu do not present an issue and are working just fine

and according to 3rd party user reports, all this used to work until january when some changes went live on stackblitz (according to some unconfirmed notices in old issues, thats when new module parser when live)?

note: this issue is also being tracked by @tensorflow team as a P0 issue - see https://github.com/tensorflow/tfjs/issues/6315

[markwhitfeld]

@vladmandic This issue should be resolved now. Apologies for the radio silence, my PC’s SSD crashed last week! Let me know if you have any issues.

[vladmandic]

in case you need to understand tensorflow module internal dependencies:

• @tensorflow/tfjs-core core module itself without any platform dependent code (there are few others like that, -data, -layers, -converter)
• @tensorflow/tfjs-backend-cpu math ops implemented in js, basically a fallback when optimized implementation is not present; without dependencies on platform-specific implementations (thus the issue here - it does not reference or need tfjs-node)
• @tensorflow/tfjs-node core + cpu + bindings to tensorflow.so so it runs optimized for nodejs
• @tensorflow/tfjs-backend-webgl core + cpu + optimized math in webgl for browser environments
• @tensorflow/tfjs - bundle of core + data + layers + converter + cpu + webgl