Implement Jupyter integration

[xbwwj]

What's Changed

This pull request implements Jupyter integration for arrow-js, including the following two mime type:

• text/html
• application/vnd.dataresource+json

This implementation basically follows nodejs-polars's implementation.

Closes #245.

[xbwwj]

There is another point to discuss before I can implement it: how many rows and columns to display.

• pandas use the global options display.max_rows, display,max_columns, display,max_colwidth
• polars use the environment variables POLARS_FMT_MAX_COLS, POLARS_FMT_MAX_ROWS, POLARS_FMT_STR_LEN (and Config contex)

I'm not sure which one is preferable.

[xbwwj]

I also failed to run yarn build with ECONNREFUSED error. How can I run the tests?

[xbwwj]

Here is an example of this integration (can preview on GitHub):

https://gist.github.com/xubaiwang/d56db9d84c671dbaa75901813941bd3b

[domoritz]

Thanks for the pull request. I think we should not have this function live on the table (if possible) since it cannot be treeshaken. Is there a way to have it outside so we can tree shake the code when it's not needed?

[xbwwj]

@domoritz I think it's currently impossible.

Although it can be moved to a separate function, this is far less useful in Jupyter context. People do not want an extra function to inspect a single value.

Is it possible to create a Deno specific target/platform (called apache-arrow/deno or apache-arrow/jupyter)?
As far as I know, all other JavaScript kernel has discontinued (IJavascript, tslab).

If we make this feature deno/jupyter platform specific (through prototype mutation and .d.ts declare), it will not affect other platforms.

[domoritz]

Maybe it's small enough to be okay.

Should columns also be supported?

And for perf, does += perform similar to building an array and then joining it into a string?

[xbwwj]

And for perf, does += perform similar to building an array and then joining it into a string?

This is why this is still a draft, there are many things left to implement and optimize.

Should columns also be supported?

Do you mean something like pandas.Series?

Maybe it's small enough to be okay.

However, the final implementation may be non trivial. There are other things like date formatting, json schema creation (for vnd.dataresource+json MIME type), formatting options, etc.

[domoritz]

I see. So maybe it does make sense to have a separate deno build then.