feat: special case for reading std::string from a TDirectory

[jpivarski]

Fixes #1159. (That's a discussion.)

What do you think, @chrisburr? Is std::string truly exceptional, such that a special case is warranted, or is this the tip of an iceberg that should be handled in a more systematic way?

(If it's to be handled in a systematic way, I don't know when or by whom.)

[jpivarski]

Actually, this second commit is less invasive, and it's setting up less of a precedent for special cases. The serialization of the std::string is just like a TString (not TObjString), so maybe we can put it in the lookup table under both names.

The downside of this approach is that it returns the data as an instance of a class named Model_TString. Although this is a subclass of str and you can easily get at the data, it could be misleading to call this "TString", given that it's a std::string in C++.

[jpivarski]

Can we include your file as a test? Is https://cburr.web.cern.ch/string-example.root a stable address?

[chrisburr]

Thank you for following up with this so quickly. I've tried this out and it works exactly the way I'd expect:

In [1]: import uproot
   ...: import json
   ...: uproot.classes["string"] = uproot.classes["TString"]
   ...:
   ...: fn = "https://cburr.web.cern.ch/string-example.root"
   ...:
   ...: f2 = uproot.open(fn)
   ...: json.loads(f2["FileSummaryRecord"])
   ...:
Out[1]:
{'LumiCounter.eventsByRun': {'counts': {},
  'empty': True,
  'type': 'LumiEventCounter'},
 'guid': '5FE9437E-D958-11EE-AB88-3CECEF1070AC'}

I suspect string is exceptional and at a glance there seems to be dedicated code for serialising it:

https://github.com/root-project/root/blob/e7d7ee19abd630b0df3a16d84d61bb59e2e9b357/core/meta/src/TClass.cxx#L648-L724

Can we include your file as a test? Is https://cburr.web.cern.ch/string-example.root a stable address?

It can absolutely be included as a test though I'd suggest putting it somewhere else if you can as that URL likely won't last forever. It's only ~5KB.

[jpivarski]

Well, if it's a special case in ROOT, it can be a special case in Uproot.

I'm still not sure if I like the

            if self._fClassName == "string":
                return cursor.string(chunk, context)

or the

uproot.classes["string"] = Model_TString

better. The first option is "more hacky," or at least, more special of a special case, and the second is natural but has the downside of declaring that a std::vector is a TString.

Actually, that made my decision for me: we can't have Uproot insinuating that a std::vector is a TString. That kind of (deliberate!) confusion might mess with someone's debugging someday. At least the explicit if statement calls itself out as an obvious wart.