René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Replace `Transform.__add__` with `Transform.__matmul__`?","articleBody":"\u003c!--To help us understand and resolve your issue, please fill out the form to the best of your ability.--\u003e\r\n\u003c!--You can feel free to delete the sections that do not apply.--\u003e\r\n\r\n### Bug report\r\n\r\n**Bug summary**\r\n\r\nWhile working on fixing the transform pipeline for image rendering, it struck me that the operator \"+\" probably sends the wrong impression to users in terms of expected properties. The Euclidean group is barely associative, and definitely not commutative. \r\n\r\nAlso, since a `Transform` is something that's *applied* (like an operator), my expectation as I'm using composite transforms is that they will be applied rightmost-first. In other words, I keep expecting\r\n\r\n```python\r\n\u003e\u003e\u003e (A + B).transform(X) == A.transform(B.transform(X))  # but it's currently False\r\n\u003e\u003e\u003e (A + B).transform(X) == B.transform(A.transform(X))  # how it currently works.\r\n```\r\n\r\nNow that Python has the \"@\" operator, this feels like a more natural choice for \"combining\" `Transforms`. People don't generally expect `A@B == B@A`, while they do expect `A + B == B + A`, which is definitely not true for Transforms. \r\n\r\nThe implementation would just be something like (modulo deprecation code)\r\n\r\n```python\r\ndef __matmul__(self, other):\r\n    return other + self\r\n```\r\n\r\nAnd we would get \r\n```python\r\n\u003e\u003e\u003e (A @ B).transform(X) == A.transform(B.transform(X))  # returns True\r\n```\r\n\r\n**Release strategy** \r\n\r\nFixing all our internal uses is (probably) only one PR's worth of work, since (relative to how much code would churn) it would be easy to verify/review. \r\n\r\nHowever, to me this feels like an integral enough part of the Transform API that it might make sense to leave in \"+\" until 4.x, and maybe just make sure we only advertise \"@\" in e.g. the tutorials?\r\n\r\nIn summary:\r\n\r\n**Pros**\r\n1. \"@\" implies lack of commutivity, whereas \"+\" implied commutivity\r\n2. \"@\" implies order of application (rightmost-first), whereas \"+\" is ambigious\r\n\r\n**Cons**\r\n1. Pretty core part of `Transform` API\r\n2. Lots of code-churn in `Transform`-related internals.\r\n\r\n\r\n\r\n","author":{"url":"https://github.com/brunobeltran","@type":"Person","name":"brunobeltran"},"datePublished":"2020-04-23T21:19:11.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":6},"url":"https://github.com/17230/matplotlib/issues/17230"}