Schemapi debug mode always on?

[velochy]

I was debugging some performance issues in my streamlit application and noticed that Altair schema validation is taking a significant amount of rendering time.

So I dug into the altair code and noticed something rather odd in altair/utils/schemapi.py, row 23-33

DEBUG_MODE = True


def enable_debug_mode():
    global DEBUG_MODE
    DEBUG_MODE = True


def disable_debug_mode():
    global DEBUG_MODE
    DEBUG_MODE = True

So if I read it correctly, debug mode is set by default, and "disable_debug_mode()" will fail to disable it.

Its hard to believe this is intended behavior, but I'd prefer someone who understand the codebase better have a look.

[binste]

Thanks for reporting this! Having DEBUG_MODE = True is actually the intended behaviour when it was introduced in #546, see that PR for the reasoning but it's about providing more informative error messages. I think the name DEBUG_MODE is maybe a bit unfortunate.

disable_debug_mode should definitely set it to False, that's a bug which should be fixed.

I proposed a rework of the validation over at #2842 and I'm wondering if we still need this validation upon creation of objects. The error messages are now informative without it so that's no longer a relevant concern but it provides error messages sooner when you create certain objects before you display them which can be nice. @velochy Do you have any performance measurements you can share? I'd very much appreciate this to get a better understanding of how much this feature slows users down. What do you mean with "significant amount of rendering time"? How does this compare to setting DEBUG_MODE = False?

[binste]

I used the %%timeit cell magic in Jupyter to time the following example. With DEBUG_MODE = True it takes 12.2 ms ± 499 µs per loop (mean ± std. dev. of 7 runs, 100 loops each). If set to False, I get 8.22 ms ± 250 µs per loop (mean ± std. dev. of 7 runs, 100 loops each). This is a difference which should not be noticeable for a user.

stocks = data.stocks.url
highlight = alt.selection_point(name="highlight", on="mouseover", clear="mouseout")

line = (
    alt.Chart(height=100, width=200)
    .mark_line(point=True)
    .encode(x=alt.X("date:N", timeUnit="year"), y=alt.Y("mean(price):Q"))
)

rect = (
    alt.Chart(width=200)
    .mark_rect()
    .encode(
        x=alt.X("date:T", timeUnit="year"),
        fill=alt.Y("mean(price):Q"),
        stroke=alt.condition(
            highlight, alt.value("black"), alt.value(None), empty=False
        ),
    )
    .add_params(highlight)
)

alt.vconcat(line, rect, data=stocks)

The advantage of the immediate validation with DEBUG_MODE = True is that errors can be catched earlier. If I for example introduce an argument unknown to selection_point I'd immediately get the correct error without having to wait for the chart to be rendered:

highlight = alt.selection_point(name="highlight", on="mouseover", clear="mouseout", unknown="test")

altair.vegalite.v5.schema.core.PointSelectionConfig, validating 'additionalProperties'

        Additional properties are not allowed ('unknown' was unexpected)

[mattijn]

I think you mean #2842 and not #2824 for the rework of the validation.

[binste]

Indeed, thanks! Edited my comment.