geom_bar (stacked bars) is slow

First, thanks for the nice package.

I was working on a problem and noticed the plotnine implementation of geom_bar is slow especially for stacked bar plots. Below is an example:

from plotnine import ggplot, geom_bar, aes
import pandas as pd
import random

n = 4000
data = pd.DataFrame(
    {
        "a": range(0, n),
        "u": random.sample(range(1, 2 * n), n),
        "v": random.sample(range(1, 2 * n), n),
        "w": random.sample(range(1, 2 * n), n),
        "x": random.sample(range(1, 2 * n), n),
        "y": random.sample(range(1, 2 * n), n),
    }
)

data1 = pd.melt(data, id_vars=["a"], var_name="sty", value_name="value")
p = ggplot() + geom_bar(aes(x="a", weight="value", fill="sty"), data=data1)

%timeit p.save("test.pdf")

On my machine, it takes about 30 seconds to make a plot. Yet, the similar code in R would takes only about 5 seconds. The R code is attached below.

library(ggplot2)
library(data.table)
library(rbenchmark)

n = 4000
data = data.table(a=1:n,
                  u=runif(n), v=runif(n), w=runif(n), x=runif(n), y=runif(n))

data1 = melt(data, id.vars="a", variable.name="sty", value.name="value")

p = ggplot() + geom_bar(aes(x=a,weight=value,fill=sty), data=data1)

benchmark(ggsave("test1.pdf", p), replications=10)

Start by reproducing the Python benchmark in the issue with geom_bar and p.save("test.pdf"), then trace the geom_bar rendering path involved in stacked bars. Compare the result with the reported roughly 30-second runtime and identify where time is spent; done means materially improving stacked-bar plot generation while preserving the expected output.