[guitargeek]

I suspect the tutorial was only JIT-heavy because in the original cppjit version in ROOT, it didn't cache some of the wrapper code that can be generated. This situation was improved by the recent improvements in cppjit.

This reverts commit 4e1572c.

FYI @aaronj0

[dpiparo]

Thanks. Before proceeding with the merge, could perhaps the timing be crosschecked one last time?

[guitargeek]

@dpiparo, sure! Here is the timings for Windows x64 and x86 before and after cppjit (where the "after cppjit" for x86 is from this PR because it didn't get run before):

Windows x64 before cppjit (6.40 nightlies):

tutorial-math-fit-fitNormSum-py ................   4.94 sec
tutorial-math-fit-fitConvolution-py ............   8.03 sec

Windows x64 now (master nightlies):

tutorial-math-fit-fitNormSum-py ................    5.77 sec
tutorial-math-fit-fitConvolution-py ............    9.22 sec

Windows x86 before cppjit (6.40 nightlies):

tutorial-math-fit-fitNormSum-py ................    6.32 sec
tutorial-math-fit-fitConvolution-py ............   10.65 sec

Windows x86 now (this PR):

tutorial-math-fit-fitNormSum-py ................   10.57 sec
tutorial-math-fit-fitConvolution-py ............   11.59 sec

Note that these measurements are quite flaky because of the concurrent ctest. But it seems that these tests in particular didn't seem to have benefited from cppjit, likely because of the fallback to string manipulation and wrapper code generation for the function callback. But at least the timing is in the same ballpark, especially for the x64 nightlies before and after cppjit. So I'd go ahead with this PR!

[github-actions]

Test Results

    23 files      23 suites   3d 22h 13m 54s ⏱️
 3 883 tests  3 883 ✅ 0 💤 0 ❌
79 827 runs  79 827 ✅ 0 💤 0 ❌

Results for commit 5f86c96.

[guitargeek]

/backport to 6.42

[root-project-bot]

Preparing to backport PR #23930 to branch 6.42 requested by guitargeek

[root-project-bot]

This PR has been backported to branch 6.42: #23934