Remove redefinition of tensorexpr

[Yue-Zhengyuan]

EDIT: This PR requires MPSKit v0.13.7 or higher.

---

This PR removes the following redefinition of MPSKit.tensorexpr:

PEPSKit.jl/src/algorithms/contractions/localoperator.jl

Lines 4 to 11 in f482711

	# currently need this because MPSKit restricts tensor names to symbols
	_totuple(t) = t isa Tuple ? t : tuple(t)
	MPSKit.tensorexpr(ex::Expr, inds::Tuple) = Expr(:ref, ex, _totuple(inds)...)
	function MPSKit.tensorexpr(ex::Expr, indout, indin)
	return Expr(
	:typed_vcat, ex, Expr(:row, _totuple(indout)...), Expr(:row, _totuple(indin)...)
	)
	end

When I look at the current MPSKit v0.13.6 (actually, since v0.12.6 or QuantumKitHub/MPSKit.jl#256), MPSKit no longer "restricts tensor names to symbols":

https://github.com/QuantumKitHub/MPSKit.jl/blob/440acdb19fb12b7b9e45b2b02ad865ab8adb66a1/src/utility/utility.jl#L123-L128

tensorexpr(name, inds) = Expr(:ref, name, _totuple(inds)...)
function tensorexpr(name, indout, indin)
    return Expr(
        :typed_vcat, name, Expr(:row, _totuple(indout)...), Expr(:row, _totuple(indin)...)
    )
end

Since now PEPSKit depends on MPSKit v0.13 or higher, we should be able to remove the redefinition of tensorexpr safely.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/algorithms/contractions/localoperator.jl	95.29% <ø> (+0.31%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[lkdvos]

Seems like there are still some corner cases left that were not caught by the MPSKit implementation. Do you want to add them to a PR there or just leave this as is for now? In principle this is not a big copy, and tensorexpr is kind of not marked as public in MPSKit anyways

[Yue-Zhengyuan]

All the errors seem to come from reduced_densitymatrix, complaining length(::Symbol) is not defined. It may still be worthwhile understanding what happened...

[lkdvos]

The problem is that tensorexpr(:name, :i, :j) is not caught as one of the edge cases, as the MPSKit version expects tensorexpr(:name, (:i,), (:j,)) instead, and while trying to canonicalize everything, it tries to convert the Symbol into a tuple, calling Tuple(::Symbol) however doesn't do the trick, which is where the error is coming from. You can see that I solved this in PEPSKit with _totuple which does a better job at canonicalizing all of the inputs. There are thus three obvious solutions:

1. Adapt the MPSKit implementation to use something similar to _totuple so that this can be caught
2. Adapt the generated functions to call tensorexpr(:name, (:i,), (:j,)) directly instead.
3. Just leave everything as is 🙃

[Yue-Zhengyuan]

There actually is a _totuple defined in MPSKit, which differs from the one in PEPSKit only by (the case of) one letter 😅:

• PEPSKit: _totuple(t) = t isa Tuple ? t : tuple(t)
• MPSKit: _totuple(t) = t isa Tuple ? t : Tuple(t)

Then I think we should modify MPSKit.

[Yue-Zhengyuan]

There are some weird segmentation faults in the tests which are also encountered in #273, so they are not caused by the PR, but some upstream regressions.

[leburgel]

It seems like it could be an issue with Julia v1.12.0 itself. I can't see much difference in package versions at first sight, the differences from the last time CI passed is that Julia v1.12.0 is used versus v1.11.7 before, and along with that LinearAlgebra.jl is on v1.12.0 versus v1.11.7 before. Most likely Julia v1.12.0 broke Zygote?

[lkdvos]

@kshyatt is this what you meant with the thing caused by GC pressure?

[Yue-Zhengyuan]

Might be related to JuliaLang/julia#59138.