fix another bug in circular type detection

[vtjnash]

Similar to #35275, but through a supertype instead of through the parameters
Fixes #41503
Fixes #41349

[JeffBezanson]

Suggested change

	// check passes, then we also still need to check the object
	// check passes, then we also still need to check the fields too.

[JeffBezanson]

Replaces #41359 ?

[Sacha0]

Regarding #41503, we hit what appears to be that issue while testing 1.7-beta3. Carrying this patch regrettably does not seem to resolve the presentation that we hit, though reverting 7ffc10b does. Hope that's helpful :).

[Sacha0]

(Please find a stacktrace in #41503 (comment).)

[JeffBezanson]

@Sacha0 can you provide an example of a case that still fails with this PR?

[Sacha0]

@JeffBezanson I will try to reduce the issue enough to pass something useful along tomorrow. Would an rr trace do the trick, absent an MRE? :)

[antoine-levitt]

The issue in #41532 is not fixed by this patch for me (tested in nightly). Reproducer is adding and testing DFTK.

[JeffBezanson]

Would an rr trace do the trick, absent an MRE? :)

An rr trace is as good as gold, as always :)

[antoine-levitt]

When running under --bug-report=rr I got a segfault ending with

┌ Error: Debugged process failed
│   exitcode = 0
│   termsignal = 11
└ @ BugReporting ~/.julia/packages/BugReporting/AZD4C/src/BugReporting.jl:191
Segmentation fault (core dumped)

with no prompt to upload a trace. There's 420M worth of files in a folder in .julia/artifacts/, should I upload that somewhere? But really Pkg.add("DFTK"); Pkg.test("DFTK") reproduces the bug reliably and so should be easier to debug than rr.

[Sacha0]

Apologies for not noting the following in this thread earlier: We shipped an rr trace to relevant folks out-of-band, from which they were able to produce an MRE, so there may be no need for more at the moment :).