html_url,issue_url,id,node_id,user,created_at,updated_at,author_association,body,reactions,issue,performed_via_github_app https://github.com/simonw/sqlite-utils/pull/333#issuecomment-974754412,https://api.github.com/repos/simonw/sqlite-utils/issues/333,974754412,IC_kwDOCGYnMM46GZJs,9599,2021-11-21T04:35:32Z,2021-11-21T04:35:32Z,OWNER,"Some other recent projects (like trying to get this library to work in JupyterLite) have made me much more cautious about adding new dependencies, especially dependencies like `pyarrow` which require custom C/Rust extensions. There are a few ways this could work though: - Have this as an optional dependency feature - so it only works if the user installs `pyarrow` as well - Implement this as a separate tool, `parquet-to-sqlite` - which could itself depend on `sqlite-utils` - Add a concept of ""plugins"" to `sqlite-utils`, similar to how those work in Datasette: https://docs.datasette.io/en/stable/plugins.html My favourite option is `parquet-to-sqlite` because that can be built without any additional changes to `sqlite-utils` at all! I find the concept of plugins for `sqlite-utils` interesting. I've so far not had quite enough potential use-cases to convince me this is worthwhile (especially since it should be very easy to build out separate tools entirely), but I'm ready to be convinced that a plugin mechanism would be worthwhile.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",1039037439, https://github.com/simonw/sqlite-utils/pull/333#issuecomment-979345527,https://api.github.com/repos/simonw/sqlite-utils/issues/333,979345527,IC_kwDOCGYnMM46X6B3,2118708,2021-11-25T16:31:47Z,2021-11-25T16:31:47Z,NONE,"Thanks for your reply @simonw . Tbh, my first attempt was actually the `parquet-to-sqlite` package but I already had Makefiles that relied on `SQLite-utils` and it was less intrusive to my workflow. Maybe I'll revisit that decision. FYI: there's a `[sqlite-parquet-vtable](https://github.com/cldellow/sqlite-parquet-vtable)` I don't think plugins make much sense either. Probably defeats the purpose of simplicity: simple database along with a pip-able package.","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",1039037439, https://github.com/simonw/sqlite-utils/pull/333#issuecomment-979442854,https://api.github.com/repos/simonw/sqlite-utils/issues/333,979442854,IC_kwDOCGYnMM46YRym,9599,2021-11-25T19:47:26Z,2021-11-25T19:47:26Z,OWNER,"I just remembered that there's one other place that this could fit: as a Datasette ""insert"" plugin. This is vaporware at the moment, but the idea is that Datasette itself could grow a mechanism for importing data, that's driven by plugins. Out of the box Datasette would be able to import CSV and CSV files, similar to `sqlite-utils insert ... --csv` - but plugins would then be able to add support for additional format such as GeoJSON or - in this case - Parquet. The neat thing about having it as a Datasette plugin is that one plugin would enable three different ways of importing data: 1. Via a new `datasette insert ...` CLI option (similar to `sqlite-utils`) 2. Via a web form upload interface, where authenticated Datasette users would be able to upload files 3. Via an API interface, where files could be programatically submitted to a running Datasette server I started fleshing out this idea quite a while ago but didn't make much concrete progress, maybe I should revisit it: - https://github.com/simonw/datasette/issues/1160","{""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",1039037439,