Feature: Reusable switching functions

[Schlevidon]

Motivation

Generating switching functions during integration is relatively expensive (complex mtree manipulation, exporting to file, etc.). In many cases, the same switching function will be generated several times (e.g. when integrating the same RHS multiple times with different parameters for optimization). To improve performance when encountering a switch, we could instead check if the corresponding switching function already exists before creating a new one from scratch.

Changes

• Fully refactored the process used to generate switching functions.
  • Completely separated the interface from the datahandle to make it significantly easier to use and test.
  • Added ability to correctly handle RHS which contain recursive helper functions.
  • Improved readability, maintainability and code quality in general.
• Introduced a new signature class, which can be used to uniquely identify a switching function based on the switch_condition, ctrlif_index and function_index.
• Naming of switching functions is now based on the hash of the signature, which makes it possible to detect if a switching function already exists and can be reused.
  • Although hash collisions are extermely unlikely, they are handled appropriately by adding a collision index to the name if necessary.
• When encountering a switch, IFDIFF will first compute the signature and hash, search the switching function directory to determine if this function already exists and finally, either reuse an existing function or create a new function from scratch.
  • The same mechanism also applies to jump functions.
  • This can be disabled in the config, e.g. for running tests or debugging.

[Schlevidon]

Note for review: Tests for the new functionality are not added yet, but will be before merging this pull requests.
May start the review already, since the missing tests don't inhibit the review process for the rest of the code.

[MichaelStrik]

Great work, the code is much more readable, well structured and I could confirm a significant reduction in runtime.
I suggest minor cosmetic changes that you can merge from pull request #33 if you agree.

[andreassommer]

Consider: Caller -> idxMtreeCaller Callee -> idxMtreeCallTarget Am 2025-05-13 11:53, schrieb Lev Gromov:

@Schlevidon commented on this pull request. ------------------------- On GenerateSwitchingFunctions/FunctionIterator.m [1]: I had considered using those variable names originally, but later on decided against it, because the names idxMtreeCaller and idxMtreeCallee only differ in a single letter, which makes it very easy to mix up these two. I think idxMtreeCall is a decent compromise in being descriptive enough, but also not hindering readability. But alternative suggestions are very welcome as always! -- Reply to this email directly, view it on GitHub [1], or unsubscribe [2]. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

Links: ------ [1] #29 (comment) [2] https://github.com/notifications/unsubscribe-auth/AHCI6K5QYMQ45IKWJOQUU2L26G6LPAVCNFSM6AAAAAB2GWFHKWVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDQMZWGA2TCMJZGI