The first conclusion is that meeting the main goal of this project (i.e., improving the performance of
& related methods) has been more difficult than expected; as a result of this unplanned-complexity, quite a few relevant outputs have been generated. Examples: various open-source applications (i.e., ParseNumber_Test.exe
), good insights into the best methodologies to accurately measure time differences between similar programs or more knowledge about performance improvements in unmanaged-C# codes. In general, this has been a very good first experience in open .NET (also my first open-source contributions ever).
Just by looking at the original goal of this project (i.e., more efficient version to be pulled to CoreCLR), the conclusion is that the final result has also been very good: the modified code is clearly quicker (i.e., every time and under all the testing conditions) without affecting the original algorithm at all and without adding any negative issue (e.g., lower readability or scalability).
The whole evolution of this project has also been a good mood-booster (and self-promotion). I chose the first chunk of code I saw in CoreCLR (by pure accident, while pre-analysing my original
-related concerns, #2285
), which happened to be in one of the most basic, old and highly-optimised parts of the whole .NET Framework (i.e., numeric type parsing inside
). I started this optimisation project even before having analysed the code in depth. Everything gets more complicated than expected, what makes me spend here much more time and effort than originally planned (by bearing in mind that this is a R&D, no-income-generating activity). And as a result of all this, I deliver what, in my opinion, is the most interesting & comprehensive project about programming since this site was created!
As explained in similar updates in other sections, this PR was accepted and even provoked modifications in other repositories (i.e., CoreRT and CoreFX versions of the same methods). Although it took quite long (over 8 months), I understand that modifications affecting a so essential part (i.e., used by the
methods of all the numeric types) cannot be accepted right away. I am certainly happy with how everything went.