Some tests other than mathtext also use these symbols, for example:

lib/matplotlib/tests/test_text.py::test_multiline2:

I am confused as to why the mathfont tests (which do not involve subsuper) have baseline changes? (perhaps I'm missing something obvious...)

It looks like maybe you updated mathfont_* when only mathtext_* were necessary? Also, maybe 18 doesn't need an update, but it's hard to tell from the GitHub diff only.

There's a good reason to believe 18 shouldn't be updated, but I think the global positioning inside svg is changing?
This is the failed diff:

(And yes, my bad, I just used the IDs of the baselines - so it included mathtext_* and mathfont_*)
Thanks for pointing it out @anntzer @QuLogic :)

Yes, I can see that 18 needs updates in some places but not all of them. Try using tools/triage_tests.py to see what ones really need updates.

some places but not all of them

That's right, and it was deliberate - for 18 only these fail (6 out of 15 variants):

18-pdf-mathtext-cm
18-pdf-mathtext-dejavuserif
18-pdf-mathtext-stix
18-svg-mathtext-stix
18-svg-mathtext-stixsans

For IDs like 22 or 67, (15 out of 15 fail):

67-pdf-mathtext-cm
67-pdf-mathtext-dejavusans
67-pdf-mathtext-dejavuserif
67-pdf-mathtext-stix
67-pdf-mathtext-stixsans
67-png-mathtext-cm
67-png-mathtext-dejavusans
67-png-mathtext-dejavuserif
67-png-mathtext-stix
67-png-mathtext-stixsans
67-svg-mathtext-cm
67-svg-mathtext-dejavusans
67-svg-mathtext-dejavuserif
67-svg-mathtext-stix
67-svg-mathtext-stixsans

I believe we're comparing images with a tolerance?
Which is why I thought it's better to update all variants for IDs like 18.

We could change just the failing images and forget about the IDs, it should not raise an error. Should I update the PR?

I believe we're comparing images with a tolerance?

No, the default tolerance is 0, and I don't see the mathtext tests setting anything different.

We could change just the failing images and forget about the IDs, it should not raise an error. Should I update the PR?

Yes, better to reduce test image churn when there is no failure necessitating an update.

With the latest push only 67 failing mathtext images are updated, regardless of their IDs.

to_rewrite = [
  "mathtext_cm_18.pdf",
  "mathtext_dejavuserif_18.pdf",
  "mathtext_stix_18.pdf",
  "mathtext_stix_18.svg",
  "mathtext_stixsans_18.svg",
  "mathtext_cm_22.pdf",
  "mathtext_dejavusans_22.pdf",
  "mathtext_dejavuserif_22.pdf",
  "mathtext_stix_22.pdf",
  "mathtext_stixsans_22.pdf",
  "mathtext_cm_22.png",
  "mathtext_dejavusans_22.png",
  "mathtext_dejavuserif_22.png",
  "mathtext_stix_22.png",
  "mathtext_stixsans_22.png",
  "mathtext_cm_22.svg",
  "mathtext_dejavusans_22.svg",
  "mathtext_dejavuserif_22.svg",
  "mathtext_stix_22.svg",
  "mathtext_stixsans_22.svg",
  "mathtext_dejavuserif_29.pdf",
  "mathtext_dejavusans_29.svg",
  "mathtext_cm_34.pdf",
  "mathtext_dejavusans_34.pdf",
  "mathtext_dejavuserif_34.pdf",
  "mathtext_stix_34.pdf",
  "mathtext_stixsans_34.pdf",
  "mathtext_cm_34.png",
  "mathtext_dejavusans_34.png",
  "mathtext_dejavuserif_34.png",
  "mathtext_stix_34.png",
  "mathtext_stixsans_34.png",
  "mathtext_cm_34.svg",
  "mathtext_dejavusans_34.svg",
  "mathtext_dejavuserif_34.svg",
  "mathtext_stix_34.svg",
  "mathtext_stixsans_34.svg",
  "mathtext_cm_52.pdf",
  "mathtext_dejavusans_52.pdf",
  "mathtext_dejavuserif_52.pdf",
  "mathtext_stix_52.pdf",
  "mathtext_stixsans_52.pdf",
  "mathtext_cm_52.png",
  "mathtext_dejavusans_52.png",
  "mathtext_dejavuserif_52.png",
  "mathtext_stix_52.png",
  "mathtext_stixsans_52.png",
  "mathtext_cm_52.svg",
  "mathtext_dejavusans_52.svg",
  "mathtext_dejavuserif_52.svg",
  "mathtext_stix_52.svg",
  "mathtext_stixsans_52.svg",
  "mathtext_cm_67.pdf",
  "mathtext_dejavusans_67.pdf",
  "mathtext_dejavuserif_67.pdf",
  "mathtext_stix_67.pdf",
  "mathtext_stixsans_67.pdf",
  "mathtext_cm_67.png",
  "mathtext_dejavusans_67.png",
  "mathtext_dejavuserif_67.png",
  "mathtext_stix_67.png",
  "mathtext_stixsans_67.png",
  "mathtext_cm_67.svg",
  "mathtext_dejavusans_67.svg",
  "mathtext_dejavuserif_67.svg",
  "mathtext_stix_67.svg",
  "mathtext_stixsans_67.svg"
  ]

I think this is supposed to go with the 3.4 release @QuLogic
Right @anntzer?

It would need a second review.

Just a few points regarding the baseline images changes:

• I would just delete test 18, because the math expression is strictly included in the expression of test 22.
• I would group the changed tests together into single expressions (resulting in a single test image), because not having to repeatedly embed the glyphs in each of the pdf/svg files results in significant savings. IOW, the following patch

diff --git i/lib/matplotlib/tests/test_mathtext.py w/lib/matplotlib/tests/test_mathtext.py
index 16ac7c7ba..b5fd906d2 100644
--- i/lib/matplotlib/tests/test_mathtext.py
+++ w/lib/matplotlib/tests/test_mathtext.py
@@ -31,11 +31,13 @@ math_tests = [
     r'$x^2$',
     r'$x^2_y$',
     r'$x_y^2$',
-    r'$\prod_{i=\alpha_{i+1}}^\infty$',
+    (r'$\sum _{\genfrac{}{}{0}{}{0\leq i\leq m}{0<j<n}}f\left(i,j\right)'
+     r'\mathcal{R}\prod_{i=\alpha_{i+1}}^\infty a_i \sin(2 \pi f x_i)'
+     r"\sqrt[2]{\prod^\frac{x}{2\pi^2}_\infty}$"),
     r'$x = \frac{x+\frac{5}{2}}{\frac{y+3}{8}}$',
     r'$dz/dt = \gamma x^2 + {\rm sin}(2\pi y+\phi)$',
     r'Foo: $\alpha_{i+1}^j = {\rm sin}(2\pi f_j t_i) e^{-5 t_i/\tau}$',
-    r'$\mathcal{R}\prod_{i=\alpha_{i+1}}^\infty a_i \sin(2 \pi f x_i)$',
+    None,
     r'Variable $i$ is good',
     r'$\Delta_i^j$',
     r'$\Delta^j_{i+1}$',
@@ -47,7 +49,7 @@ math_tests = [
     r"$f'\quad f'''(x)\quad ''/\mathrm{yr}$",
     r'$\frac{x_2888}{y}$',
     r"$\sqrt[3]{\frac{X_2}{Y}}=5$",
-    r"$\sqrt[5]{\prod^\frac{x}{2\pi^2}_\infty}$",
+    None,
     r"$\sqrt[3]{x}=5$",
     r'$\frac{X}{\frac{X}{Y}}$',
     r"$W^{3\beta}_{\delta_1 \rho_1 \sigma_2} = U^{3\beta}_{\delta_1 \rho_1} + \frac{1}{8 \pi 2} \int^{\alpha_2}_{\alpha_2} d \alpha^\prime_2 \left[\frac{ U^{2\beta}_{\delta_1 \rho_1} - \alpha^\prime_2U^{1\beta}_{\rho_1 \sigma_2} }{U^{0\beta}_{\rho_1 \sigma_2}}\right]$",
@@ -89,11 +91,13 @@ math_tests = [
     r'${x}_{92}^{31415}+\pi $',
     r'${x}_{{y}_{b}^{a}}^{{z}_{c}^{d}}$',
     r'${y}_{3}^{\prime \prime \prime }$',
+    # End of the MathML torture tests.
+
     r"$\left( \xi \left( 1 - \xi \right) \right)$",  # Bug 2969451
     r"$\left(2 \, a=b\right)$",  # Sage bug #8125
     r"$? ! &$",  # github issue #466
     None,
-    r'$\sum _{\genfrac{}{}{0}{}{0\leq i\leq m}{0<j<n}}P\left(i,j\right)$',
+    None,
     r"$\left\Vert a \right\Vert \left\vert b \right\vert \left| a \right| \left\| b\right\| \Vert a \Vert \vert b \vert$",
     r'$\mathring{A}  \AA$',
     r'$M \, M \thinspace M \/ M \> M \: M \; M \ M \enspace M \quad M \qquad M \! M$',

(together with updating image 18 for the new test) shrinks the overall size of the new baselines from 508K to 256K, a not insignificant saving. (I explicitly left test 52 alone as we can reasonably leave the "mathml torture test" as is; I also slightly tweaked the mathtext expression to reduce the number of independent glyphs used.)

I notice something interesting:
Following two are png and svg outputs of test 52 dejavusans font.
Note how the b character is shifted, as well as the bottom i, j, k characters in the png vs svg output.

(side by side)

Maybe it's hard to notice here, but there is definitely something fishy w.r.t. png vs svg backend?

Without looking carefully at this, it may be related to #19261 (comment) / #5414? (If so, this may change depending on rcParams["text.hinting"].)

You will need to rebase if you want to fix the docs CI build. @anntzer is there anything else left to do here?

Nope, lgtm.
@QuLogic did you want docs to build?

No, it doesn't look like this has any relevant doc changes.