ManimCE 0.20.1: MathTex bounding box changes size when shifted

[uwezi]

Description of bug / unexpected behavior

When shifting a MathTex() object which has an argument in the form of a list of substrings, the bounding box of the object grows, both vertically and horizontally.

As reported by TomBC on Discord https://discord.com/channels/1453870851807117363/1482938671190511748/1482938671190511748

Expected behavior

The bounding box should not change and just be the bounding box of the object.

How to reproduce the issue

Code for reproducing the problem

class TryMathTex(Scene):
    def construct(self):
        li = ['\\overrightarrow{', 'C', 'A', '}', '=', '\\bold{', 'm', '}']
        # str = r"\overrightarrow{BD}=\bold{m}"
        math_tex = MathTex(*li) 
        self.add(math_tex)
        bg1 = BackgroundRectangle(math_tex, color=GRAY, fill_opacity=0.5)
        self.play(Create(bg1))                    
        print(math_tex.width, math_tex.height) # 1.870 0.606 when I use li to MathTex
        self.play(math_tex.animate.shift(RIGHT+UP))
        print(math_tex.width, math_tex.height) # 1.935 1.303 when I use li to MathTex
        bg2 = BackgroundRectangle(math_tex, color=GRAY, fill_opacity=0.5)
        self.play(Create(bg2))

Additional media files

Images/GIFs

Logs

Terminal output

Bounding box before .shift()

1.8702728500000008 0.60589355

Bounding box after .shift()

1.9351364250000005 1.302946775000001

System specifications

System Details

• OS: Windows 11 (my system) and Linux (Manimator)
• Python version: 3.12.8

LaTeX details

• LaTeX distribution: MikTex (my system) and TexLive (Manimator)

Additional comments

[BraverChiang]

I'm TomBC on Discord, Thank you so much for this bug report @uwezi . To help contributors of manimCE to navigate the bug a little bit faster, I post what I find below:
This shift bounding box bug can date back to PR#3423 or Commit b7eefca. When great contributors tried to Fix incorrect submobject count of multi-part Tex/MathTex, and removed this line sub_tex_mob.submobjects = [VectorizedPoint()] in def _break_up_by_substrings(self): in MathTex class in Tex_Mobject.py, which unfortunately gave rise to the shift bounding box bug.

So now, for me, or in case for someone who may need to fix the bug temporarily, I have to add manually this line back sub_tex_mob.submobjects = [VectorizedPoint()] for manim 0.18.0~0.19.2, and put this line from manim.mobject.types.vectorized_mobject import VectorizedPoint back at the top of the file Tex_Mobject.py .
And for manim 0.20.0 or later, because the MathTex class has been modified a lot, AI give me a solution as I mentioned in discord, by replacing

        try:
            for tex_string, tex_string_id in self._main_matches:
                mtp = MathTexPart()
                mtp.tex_string = tex_string
                mtp.add(*self.id_to_vgroup_dict[tex_string_id].submobjects)
                new_submobjects.append(mtp)
        except KeyError:
            logger.error(
                f"MathTex: Could not find SVG group for tex part '{tex_string}' (id: {tex_string_id}). Using fallback to root group."
            )
            new_submobjects.append(self.id_to_vgroup_dict["root"])

with

        for tex_string, tex_string_id in self._main_matches:
            mtp = MathTexPart()
            mtp.tex_string = tex_string

            vgroup = self.id_to_vgroup_dict.get(tex_string_id)

            if vgroup is None or len(vgroup.submobjects) == 0:
                # Equivalent to the old VectorizedPoint workaround
                mtp.add(VectorizedPoint())
            else:
                
mtp.add(*vgroup.submobjects)

            new_submobjects.append(mtp)

[Chessing234]

PR #4877: empty MathTexPart submobjects now get a VectorizedPoint placeholder so the bbox stays fixed after shift.

[Chessing234]

Closed #4877 — VectorizedPoint workaround breaks test_multi_part_tex_with_empty_parts and the shift bbox regression doesn't reproduce on current main (Cairo + OpenGL). Happy to revisit if there's a minimal repro on latest.