Vectorized linear function

[antonmarsden]

@lovell This looked easy enough to implement, and adds some good value. This pull request is not really production quality and my knowledge of the code base is very limited. A few things to consider:

• do the additional variables make sense (linearVecA/linearVecB) or could linearA/linearB be both number & number[]?
• are there other functions that can be generalised in a similar fashion?
• probably needs a good test case :)
• unclear what the behaviour is (or should be) if the arrays are different lengths or if only one array is defined

[lovell]

Hi, thanks for the PR.

How about we switch the (internal) this.options so linearA and linearB are always arrays, and if someone provides a single value (the current behaviour) we convert to a single element array.

.linear(a, b)
// would be equivalent to
.linear([a], [b])
// but now we would also support
.linear([a1, a2, a3], [b1, b2, b3])

This should make the C++ code path a bit simpler as it will only deal with arrays of doubles.

[antonmarsden]

@lovell Great suggestion, I have implemented it. Added some fancy logic for dealing with the alpha channel that you may wish to review. Only outstanding thing is test cases, which I can hopefully make some progress on shortly.

[antonmarsden]

Handling the multi-band expansion for linear() looked tricky in the context of sharp's colourspace handling.

[lovell]

Thanks for the updates. I've left a couple of small comments/questions inline but otherwise this is good to go.

Please can you rebase against the upstream/eagle branch.

[antonmarsden]

@lovell Done, all yours.

[coveralls]

Coverage remained the same at 100.0% when pulling 0c55d0e on antonmarsden:main into dd56a96 on lovell:eagle.

[lovell]

Please can you add unit test(s) to ensure all the new logic is fully-covered - see https://coveralls.io/builds/51160601

[lovell]

Landed via commit 74e3f73 - this will be in v0.31.0, thanks again for working on this.