(DO NOT MERGE) don't allow identifiers to start with Lm (Letter, modifier)

[stevengj]

As discussed in #28441, #28494, and #34507, it is a bit awkward that identifiers can start with a category-Lm (Letter, modifier) character (such as a superscript). This PR is an experiment to see how breaking it would be to change that. (You can still have Lm characters in identifiers, just not as the first character.)

(Breaks #24567.)

Can we please run PkgEval on this PR?

[KristofferC]

@nanosoldier runtests(ALL, vs=":master")

[stevengj]

@nanosoldier runtests(ALL, vs = ":master")

[KristofferC]

Wasn't it busy running the previous tests? The tests take something like 13 hours so might have to wait for quite a while to get the result of the second run.

[JeffBezanson]

Maybe ᵀ should be an exception? I don't know.

[maleadt]

Wasn't it busy running the previous tests? The tests take something like 13 hours so might have to wait for quite a while to get the result of the second run.

Yes, this queued two of them with no way to cancel.

[ Info: [Node 1 | 2020-01-28T09:23:12.578]: received job submission with phrase RegexMatch("@nanosoldier `runtests(ALL, vs=\":master\")`")
[ Info: [Node 1 | 2020-01-28T10:57:01.591]: received job submission with phrase RegexMatch("@nanosoldier `runtests(ALL, vs = \":master\")`")

[stevengj]

Wasn't it busy running the previous tests?

There was an error message that the request was ill-formed or something like that, which is why I submitted another. But maybe that was from a previous attempt?

[KristofferC]

Ah, Nanosoldier benchmark bot and Nanosoldier Pkgeval bot are fighting of the same check "instance".

[StefanKarpinski]

Perhaps it would be helpful if nanosoldier posted immediately that it was doing something with a link to some kind of view of its progress?

[nanosoldier]

Your test job has completed - possible new issues were detected. A full report can be found here. cc @maleadt

[DilumAluthge]

Maybe ᵀ should be an exception? I don't know.

That would fix some of them, but several are failing with invalid character "ˣ". Although as far as I can tell, those are always from line 154 of AbstractTensors, so it might be sufficient to fix that package.

[simeonschaub]

It seems like this only affects AbstractTensors, SugarBLAS and EasyTranspose. The interesting thing to note here is that they all assign some kind of singleton to a Unicode modifier and then overload multiplication as a hack to get postfix operators. I believe, #34507 would allow for doing this in a much less hacky way by just defining the adjoint operator combined with a modifier.

[nanosoldier]

Your test job has completed - possible new issues were detected. A full report can be found here. cc @maleadt

[c42f]

The interesting thing to note here is that they all assign some kind of singleton to a Unicode modifier and then overload multiplication as a hack to get postfix operators.

This seems a strong signal that the breakage is worthwhile. A more detailed survey of the mentioned packages:

EasyTranspose is a trivial package which just defines ᵀ https://github.com/musm/EasyTranspose.jl/blob/master/src/EasyTranspose.jl

SugarBLAS also defines ᵀ and uses it for things like https://github.com/lopezm94/SugarBLAS.jl/blob/master/src/SugarBLAS.jl#L274

AbstractTensors defines several postfix operators - see
https://github.com/chakravala/AbstractTensors.jl/blob/31123e0c2270fcb5e88a28c64f8cc8eaa6e28c14/src/AbstractTensors.jl#L240-L247
And tests:
https://github.com/chakravala/AbstractTensors.jl/blob/31123e0c2270fcb5e88a28c64f8cc8eaa6e28c14/test/runtests.jl#L42-L46

[JeffBezanson]

From triage:

• In the abstract we think this is the right thing to do, but it's too breaking to do all of it. Consider in 2.0.
• But, we might be able to change the parsing only of modifiers following ', so e.g. x'ᵀ becomes an operator instead of multiplication. This needs more investigation to see if such cases occur in practice. Preliminary search says probably not.