uncertainties not propagated to all constraint expressions

[newville]

Description

As shown at https://groups.google.com/g/lmfit-py/c/KGrks5UBqfs/m/F2XfM286BwAJ, uncertainties are not correctly propagated to constraint expressions that include other constraint expressions.

A Minimal, Complete, and Verifiable example

import numpy as np
from lmfit.models import SineModel

model = SineModel()
params = model.make_params(amplitude=1, frequency=9.0, shift=0)
params.add("period", expr="1/frequency")
params.add("period_2a", expr="2*period")
params.add("period_2b", expr="2/frequency")
params.add("thing_1",   expr="shift + frequency")
params.add("thing_2",   expr="shift + 1/period")

np.random.seed(3)

xs = np.linspace(0, 1, 51)
ys = np.sin(7.45 * xs) + 0.01 * np.random.normal(size=xs.shape)

fit = model.fit(ys, x=xs, params=params)
print(fit.fit_report())

which shows:

[[Model]]
    Model(sine)
[[Fit Statistics]]
    # fitting method   = leastsq
    # function evals   = 25
    # data points      = 51
    # variables        = 3
    chi-square         = 0.00505054
    reduced chi-square = 1.0522e-04
    Akaike info crit   = -464.224405
    Bayesian info crit = -458.428928
    R-squared          = 0.99978830
[[Variables]]
    amplitude:  1.00048377 +/- 0.00209960 (0.21%) (init = 1)
    frequency:  7.43904407 +/- 0.00641239 (0.09%) (init = 9)
    shift:      0.00310624 +/- 0.00382291 (123.07%) (init = 0)
    period:     0.13442587 +/- 1.1587e-04 (0.09%) == '1/frequency'
    period_2a:  0.26885175 +/- 0.00000000 (0.00%) == '2*period'
    period_2b:  0.26885175 +/- 2.3175e-04 (0.09%) == '2/frequency'
    thing_1:    7.44215031 +/- 0.00372918 (0.05%) == 'shift + frequency'
    thing_2:    7.44215031 +/- 0.00382290 (0.05%) == 'shift + 1/period'
[[Correlations]] (unreported correlations are < 0.100)
    C(frequency, shift) = -0.8531

The uncertainties in period_2a and period_2b should be the same, as should the uncertainties in thing_1 and thing_2.

Version information

Python: 3.10.9 | packaged by conda-forge | (main, Feb 2 2023, 20:24:27) [Clang 14.0.6 ]
lmfit: 1.1.0, scipy: 1.10.1, numpy: 1.24.2,asteval: 0.9.29, uncertainties: 3.1.7

Link(s)

https://groups.google.com/g/lmfit-py/c/KGrks5UBqfs/m/F2XfM286BwAJ