Unfortunatly this doesn't seem to fix the issue

Does the baseline image need to be updated?

Beyond that, I wonder if OrdereredDict isn't enough for the issue if things still come in from TeX in a non-deterministic order. We may need to forcibly sort...

Yes thats true, Not completely sure what happens but this is only an issue on 2.x so presumably ti requires something which is not backported to 2.x

The real reason this fails is that #5677 wasn't backported. That changes the exception in textpath.py to catch Runtime errors too. It's not obvious from #5677 why that is needed by @mdboom might remember

The exception changed in the rewrite of the C++ extensions last year, but none of that was in the test suite (since we didn't have TeX+SVG tests until this one).

I say we just apply

diff --git a/lib/matplotlib/textpath.py b/lib/matplotlib/textpath.py
index c063c3a..b694619 100644
--- a/lib/matplotlib/textpath.py
+++ b/lib/matplotlib/textpath.py
@@ -345,7 +345,7 @@ class TextToPath(object):
                                                     1094995778)]:
                     try:
                         font.select_charmap(charmap_code)
-                    except ValueError:
+                    except (ValueError, RuntimeError):
                         pass
                     else:
                         break

here, and not backport all of #5677 -- it turns out there are few issues with it and we may end up reverting it in the long run.

@mdboom makes sense I have applied the error change here but I guess we should backport that change (16698e8) all the way to 1.5.x?

I think the other change here should go in to since this test really needs tex to run, But that is against 2.x

I agree on all points. Not sure if it matters too much how we do it -- we can always make another PR against v1.5.x with 16698e8 cherry-picked in it.