Supporting XRootD

[jedori0228]

When we use a XRootD path (useful for grid jobs) in SpectrumLoader, this file does not pass the following line, and is neglected, since stat() returns non-zero value.

sbnana/sbnana/CAFAna/Core/Utilities.cxx

Lines 539 to 540 in 5df8255

	if(stat(p.we_wordv[i], &sb) == 0)
	fileList.push_back(p.we_wordv[i]);

A XRootD file can be used if stat() is commented out (but still the wildcard will not work for XRootD path).

[cjbacchus]

There is special logic in WildcardSource to let through /pnfs/ locations (not xrootd'd locations) even if they can't be found
https://github.com/SBNSoftware/sbnana/blob/develop/sbnana/CAFAna/Core/WildcardSource.cxx

and then in FileListSource it is supposed to automatically translate pnfs paths into their xrootd equivalents.
https://github.com/SBNSoftware/sbnana/blob/develop/sbnana/CAFAna/Core/FileListSource.cxx#L98

So, all of this should work if you just let CAFAna xrootd your paths for you. Except, the line in FileListSource is commented out (it exists in the nova version) for some reason. Can you try uncommenting this and running with a /pnfs/ path to see what happens?

[cjbacchus]

Oh, hang on. We would also need a return wildcard; after the warning in WildcardSource. So could you try with both of those changes included?

[jedori0228]

Sorry for the late response. I checked uncommenting

sbnana/sbnana/CAFAna/Core/FileListSource.cxx

Line 98 in 21333ee

// loc = pnfs2xrootd(loc); // no-op for non /pnfs locations

and just using /pnfs/ path works, except that I had to also modify

sbnana/sbnana/CAFAna/Core/Utilities.cxx

Line 820 in 48c0041

loc = std::string("root://fndcagpvm01.fnal.gov:1094//pnfs/fnal.gov/usr/")+&loc.c_str()[6];

: fndcagpvm01 -> fndca1

[wesketchum]

@jedori0228 @cjbackhouse is this still an oustanding issue, or what is the status here?

[cjbacchus]

I think the best way forward is for @jedori0228 to make a PR with the two changes mentioned above.

[wesketchum]

@jedori0228 @cjbackhouse did this ever happen?

[cjbacchus]

Comment above is still relevant. That said, this fix is in cafanacore (used by the RecordSource branch), I believe. So we could get it that way too.