[HistFactory] Disable global histogram registration for HistFactory.

[hageboeck]

The HistFactory histograms are owned by the HistFactory models, and they are explicitly written to files. Therefore, they should not participate in directory registration.

Fix #18172

[github-actions]

Test Results

    19 files      19 suites   4d 11h 37m 54s ⏱️
 2 717 tests  2 716 ✅ 0 💤 1 ❌
49 832 runs  49 831 ✅ 0 💤 1 ❌

For more details on these failures, see this check.

Results for commit ec75ccc.

♻️ This comment has been updated with latest results.

[pcanal]

Suggested change

	TDirectory::TContext dirContext;
	std::unique_ptr<TFile> outFile;
	gDirectory = nullptr; // Disable global registration of histograms this file. HistFactory places them explicitly later.
	TDirectory::TContext dirContext{nullptr} // Disable global registration of histograms this file. HistFactory places them explicitly later.
	std::unique_ptr<TFile> outFile;

Since the next line does not allocated a TFile and thus does not change gDirectory

[pcanal]

However line 138:

outFile = std::make_unique<TFile>(outputFileName.c_str(), "recreate");

should probably be:

outFile = std::make_unique<TFile>(outputFileName.c_str(), "RECREATE_WITHOUT_GLOBALREGISTRATION") // note: casing does not matter.
gDirectory = nullptr;  // Make sure the files does not become the current directory.  We do not need TContext here since we have one on the outer scope.

Side note the _WITHOUT_GLOBALREGISTRATION is probably the only thing strictly necessary for the goal of this PR (i.e. since it is not registered, it is not on the list of Cleanups and thus will not be traversed on delete of the histograms).

[hageboeck]

Oh, I had no idea this option existed! That's really good to know!

[TomasDado]

Hello, the execution is slightly faster from ~260s to about 230s but it seems most of the time is still spent in the destructors

tomas@tomas-IdeaPad-5-Pro-14IAP7:~/TRExFitter$ root --version
ROOT Version: 6.35.01
Built for linuxx8664gcc on Mar 31 2025, 07:22:16
From remotes/hageboeck/histFactoryCleanups@7b0297d5e9

[hageboeck]

Hello, the execution is slightly faster from ~260s to about 230s but it seems most of the time is still spent in the destructors

Hello @TomasDado, my attempt at changing it was mostly without effect as pointed out by Philippe. I force-pushed a new version of the commit that should have larger impact. Could you try again, please?

[TomasDado]

Hello, the execution is slightly faster from ~260s to about 230s but it seems most of the time is still spent in the destructors

Hello @TomasDado, my attempt at changing it was mostly without effect as pointed out by Philippe. I force-pushed a new version of the commit that should have larger impact. Could you try again, please?

Unfortunately, it still looks very similar, running time of about 230s:

tomas@tomas-IdeaPad-5-Pro-14IAP7:~/TRExFitter$ root --version
ROOT Version: 6.35.01
Built for linuxx8664gcc on Mar 31 2025, 11:12:47
From remotes/hageboeck/histFactoryCleanups@18a27e3303