Performance issues with Tex objects

[Elaine2048]

The created Tex object is difficult to be recycled by GC, resulting in a surge in memory usage in complex logic.
I tried the Text object, which can be completely recycled, but the Tex object is almost completely unrecyclable.
Has anyone dealt with this issue?
Many thanks!

[uwezi]

The created Tex object is difficult to be recycled by GC,

Sorry, but I don't understand what you mean here...

[Elaine2048]

I'm experiencing a memory leak issue with Manim's Tex objects. Despite explicitly deleting the objects and forcing garbage collection, the Tex objects persist in memory. Here's what I've observed:
I create Tex objects in a loop, perform some operations, and immediately delete them
I explicitly call gc.collect() after every 10 objects and at the end of the loop
After processing all objects, I check memory usage and object counts
The Tex objects are still present in memory (confirmed via objgraph.by_type('Tex'))
Physical memory (RSS) and private memory (USS) show significant increases that don't return to baseline
This behavior suggests that either:
The Tex objects are not being properly garbage collected
There are reference cycles or hidden references preventing collection

[uwezi]

I guess for any developer trying to reproduce this behavior it would be very helpful if you added the code as correctly formatted text for github, not only as screenshots. Also it might be helpful (as shown in the template) if you specified the version of Python and the version of Manim that you are using, as well as possibly your operating system.

[Elaine2048]

import gc
import psutil
import objgraph
import time
import numpy as np
from manim import MathTex, config, tempconfig,Tex

# 禁用Manim的视频渲染和缓存以减少干扰
config.verbosity = "ERROR"
config.disable_caching = True
config.renderer = "opengl"  # 使用OpenGL渲染器

def get_memory_usage():
    """获取当前进程的内存使用情况"""
    process = psutil.Process()
    mem_info = process.memory_info()
    full_info = process.memory_full_info()
    
    return {
        "物理内存(RSS)": mem_info.rss / (1024 * 1024),
        "虚拟内存(VMS)": mem_info.vms / (1024 * 1024),
        "独占内存(USS)": full_info.uss / (1024 * 1024)
    }

def get_object_counts():
    """获取关键对象的计数"""
    counts = {}
    # 添加更多Manim相关对象
    for obj_type in ["Tex", "MathTex", "SVGMobject", "VMobject", "OpenGLMobject",
                     "Mobject", "Point", "Vector", "Line", "Circle",
                     "dict", "list", "tuple", "str", "float", "np.ndarray"]:
        counts[obj_type] = len(objgraph.by_type(obj_type))
    return counts

def print_memory_report(stage_name, mem_data, obj_counts):
    """打印内存报告"""
    print(f"\n{'='*40}")
    print(f"{stage_name} 内存报告")
    print(f"{'-'*40}")
    
    print("内存使用 (MB):")
    for k, v in mem_data.items():
        print(f"{k}: {v:.2f}")
    
    print("\n对象数量:")
    for k, v in obj_counts.items():
        print(f"{k}: {v}")


def create_and_destroy_tex_objects(num_objects=50):
    """创建并销毁MathTex对象（每次循环创建并立即销毁一个对象）"""
    for i in range(num_objects):
        # 创建单个MathTex对象
        tex = Tex(f"$\\text{{Test String }} {i} \\times \\sum_{{n=1}}^{{\\infty}} \\frac{{1}}{{n^2}} = \\frac{{\\pi^2}}{{6}}$")
        
        # 模拟一些操作
        tex.scale(0.5)
        
        # 立即销毁对象
        tex.clear_points() 
        tex.remove()
        del tex
        
        # 每10个对象强制一次垃圾回收
        if i % 10 == 0:
            gc.collect()
    
    # 最后强制一次完整的垃圾回收
    gc.collect()

def main():
    # 使用临时配置减少内存占用
    with tempconfig({"renderer": "opengl", "disable_caching": True}):
        # 初始内存状态
        gc.collect()
        time.sleep(1)  # 等待系统稳定
        mem_stages = {}
        obj_stages = {}
        
        # 阶段0: 初始状态
        mem_stages[0] = get_memory_usage()
        obj_stages[0] = get_object_counts()
        print_memory_report("初始状态", mem_stages[0], obj_stages[0])
        
        # 阶段1: 创建Tex对象后
        create_and_destroy_tex_objects(num_objects=50)
        time.sleep(1)  # 等待系统内存稳定

        mem_stages[1] = get_memory_usage()
        obj_stages[1] = get_object_counts()
        print_memory_report("创建MathTex对象后", mem_stages[1], obj_stages[1])
        
        # 阶段2: 强制GC后
        gc.collect()
        time.sleep(2)  # 给GC更多时间完成清理
        mem_stages[2] = get_memory_usage()
        obj_stages[2] = get_object_counts()
        print_memory_report("GC回收后", mem_stages[2], obj_stages[2])
        
        # 内存变化分析
        print("\n\n" + "="*40)
        print("内存变化分析")
        print("="*40)
        print(f"物理内存变化: {mem_stages[2]['物理内存(RSS)'] - mem_stages[0]['物理内存(RSS)']:.2f} MB")
        print(f"虚拟内存变化: {mem_stages[2]['虚拟内存(VMS)'] - mem_stages[0]['虚拟内存(VMS)']:.2f} MB")
        print(f"独占内存变化: {mem_stages[2]['独占内存(USS)'] - mem_stages[0]['独占内存(USS)']:.2f} MB")
        
        # 对象变化分析
        print("\n对象数量变化:")
        for obj_type in obj_stages[0].keys():
            change = obj_stages[2][obj_type] - obj_stages[0][obj_type]
            if change != 0:  # 只显示有变化的类型
                print(f"{obj_type}: {change:+}")
        
        # 检查是否有Tex/MathTex泄漏
        tex_leak = obj_stages[2]["Tex"] + obj_stages[2]["MathTex"] - obj_stages[0]["Tex"] - obj_stages[0]["MathTex"]
        print("\n结论:")
        if tex_leak > 0:
            print(f"⚠️ 检测到可能的内存泄漏: Tex/MathTex对象增加了 {tex_leak} 个")
        else:
            print("✅ Tex/MathTex对象被完全回收")
        
        # 显示内存变化详情
        print(f"\n物理内存变化: {mem_stages[2]['物理内存(RSS)'] - mem_stages[0]['物理内存(RSS)']:.2f} MB")
        print(f"独占内存变化: {mem_stages[2]['独占内存(USS)'] - mem_stages[0]['独占内存(USS)']:.2f} MB")

if __name__ == "__main__":
    main()

python版本

Python 3.13.3

manim版本

Manim Community v0.19.0

[OliverStrait]

Here is the issue created from this: #4327
Using the argument flag Tex( use_svg_cache = False) will remove half of memory issues, but there was still one unexpected behavior that is described in issue and how to clear this buggy cache manually.

[Elaine2048]

Thank you very much, this method is very useful, and the number of objects has been reduced by exactly half.