issue_comments

This sounds like a valuable feature for people running Datasette behind a proxy.

Yes, in some cases it is easer to use e.g. Apache's ProxyPass Directive with Unix Domain Socket like unix:/home/www.socket|http://localhost/whatever/.

This would really help with this issue: https://github.com/eyeseast/datasette-geojson/issues/7

Related: Here's an implementation of a get_metadata() plugin hook by @brandonrobertz next-LI@3fd8ce9

Here's a plugin that implements metadata-within-DBs: next-LI/datasette-live-config

How it works: If a database has a __metadata table, then it gets parsed and included in the global metadata. It also implements a database-action hook with a UI for managing config.

More context: https://github.com/next-LI/datasette-live-config/blob/72e335e887f1c69c54c6c2441e07148955b0fc9f/datasette_live_config/__init__.py#L109-L140

Here's where the plugin hook is called, demonstrating the fallback= argument:

https://github.com/simonw/datasette/blob/05a312caf3debb51aa1069939923a49e21cd2bd1/datasette/app.py#L426-L472

I'm not convinced of the use-case for passing fallback= to the hook here - is there a reason a plugin might care whether fallback is True or False, seeing as the metadata() method already respects that fallback logic on line 459?

I think you're right. I can't think of a reason why the plugin would care about the fallback parameter since plugins are currently mandated to return a full, global metadata dict.

Hmmm... that's tricky, since one of the most obvious ways to use this hook is to load metadata from database tables using SQL queries.
@brandonrobertz do you have a working example of using this hook to populate metadata from database tables I can try?

Answering my own question: here's how Brandon implements it in his datasette-live-config plugin: https://github.com/next-LI/datasette-live-config/blob/72e335e887f1c69c54c6c2441e07148955b0fc9f/datasette_live_config/__init__.py#L50-L160

That's using a completely separate SQLite connection (actually wrapped in sqlite-utils) and making blocking synchronous calls to it.

This is a pragmatic solution, which works - and likely performs just fine, because SQL queries like this against a small database are so fast that not running them asynchronously isn't actually a problem.

But... it's weird. Everywhere else in Datasette land uses await db.execute(...) - but here's an example where users are encouraged to use blocking calls instead.

Ideally this hook would be asynchronous, but when I started down that path I quickly realized how large of a change this would be, since metadata gets used synchronously across the entire Datasette codebase. (And calling async code from sync is non-trivial.)

In my live-configuration implementation I use synchronous reads using a persistent sqlite connection. This works pretty well in practice, but I agree it's limiting. My thinking around this was to go with the path of least change as Datasette.metadata() is a critical core function.

If this is a concept ACK then I will move onto fixing the tests (adding new ones) and updating the documentation for the new plugin hook.

Fair. I looked into it, it looks like it could be done, but it would be a bit ugly. I can upload and link a gist of my exploration. Click can parse a first argument while still recognizing it as a sub-command keyword. From there, the program could:
1. ignore it preemptively if it matches a sub-command
2. and/or check if a (db) file exists at the path.

It would then also need to set a shared db argument variable.

Click also makes it easy to parse arguments from environment variables. If you're amenable, I may submit a patch for only that, which would update each sub-command to check for a DB/SQLITE_UTILS_DB environment variable. The goal would be usage that looks like: DB=./convenient.db sqlite-utils [operation] [args]

There's a workaround: https://css-tricks.com/css-fix-for-100vh-in-mobile-webkit/

and a future fix: https://css-tricks.com/safari-15-new-ui-theme-colors-and-a-css-tricks-cameo/

I'm happy to grab some or all of these in this PR, if you want.

Note that if we went with a "update_metadata" hook, the hook signature would look something like this (it would return nothing):

update_metadata(
  datasette=self, metadata=metadata, key=key, database=database, table=table,
  fallback=fallback
)

The Datasette function _metadata_recursive_update(self, orig, updated) would disappear into the plugins. Doing this, though, we'd lose the easy ability to make the local metadata.yaml immutable (since we'd no longer have the recursive update).

(Putting thoughts here to keep the conversation in one place.)

I think using datasette for this use-case is the right approach. I usually have both datasette and sqlite-utils installed in the same project, and that's where I'm trying out queries, so it probably makes the most sense to have datasette also manage the output (and maybe the input, too).

It seems like both --get and --query could work better as subcommands, rather than options, if you're looking at building out a full CLI experience in datasette. It would give a cleaner separation in what you're trying to do and let each have its own dedicated options. So something like this:

# run an arbitrary query
datasette query covid.db "select * from ny_times_us_counties limit 1" --format yaml

# run a canned query
datasette get covid.db some-canned-query --format yaml

I very much want to make:
https://disinfectants.SaferOrToxic.org/disinfectants/listN
have this URL:
https://disinfectants.SaferOrToxic.org/

I'm using only one table page on the site, with no pagination. I'm not using the home page, though when I tried to move my table to the home page as mentioned above, I failed to figure out how.

I am using cloudflare, but I haven't figured out a forwarding or HTML re-write method of doing this, either.

Is there any way I can get a prettier list URL? I'm on Vercel.

(I have a wordpress site on the main domain on a separate host.)

Note, the CVEs are probably resolvable with this https://github.com/simonw/datasette/pull/1296 . My experience is that Ubuntu seems to manage these better? Though that is surprising :/

Thanks Simon!

That corner handle looks like a hamburger menu to me. Note that the default resize handle is not limited to two-way resize: http://jsfiddle.net/LLrh7Lte/

Superseded by #1321.

Do you have a list of sqlite versions you want to test against?

One cool thing I saw recently (that we started using) was using import docker within python, and then writing pytest functions which executed against the container

setup

example

The inspiration for this came from the jupyter docker-stacks

So off the top of my head, could look at building the container with different sqlite versions as a build-arg, then run tests against the containers. Just brainstorming though

I have also found that ubuntu has fewer vulnerabilities than the buster based images.

➜  ~ docker pull python:3-buster
➜  ~ trivy image python:3-buster | head                             
2021-04-28T17:14:29.313-0400    INFO    Detecting Debian vulnerabilities...
2021-04-28T17:14:29.393-0400    INFO    Trivy skips scanning programming language libraries because no supported file was detected
python:3-buster (debian 10.9)
=============================
Total: 1621 (UNKNOWN: 13, LOW: 1106, MEDIUM: 343, HIGH: 145, CRITICAL: 14)
+------------------------------+---------------------+----------+------------------------------+---------------+--------------------------------------------------------------+
|           LIBRARY            |  VULNERABILITY ID   | SEVERITY |      INSTALLED VERSION       | FIXED VERSION |                            TITLE                             |
+------------------------------+---------------------+----------+------------------------------+---------------+--------------------------------------------------------------+

I ended up using some JS as a workaround.

First, add a JS file in metadata.yaml:

extra_js_urls:
  - '/static/app.js'

then inside the script, find the blob download links and replace .blob extension in the url with .jpg and replace the links with <img/> elements.
You need to add an output formatter to serve BLOB columns as JPG. You can find the code in the first post.
<del>Replacing .blob -> .jpg might not even be necessary, because browsers only care about the mime type, so you only need to serve the binary content with the right content-type header.</del>. You need to replace the extension, otherwise the output renderer will not run.

window.addEventListener('DOMContentLoaded', () => {
    function renderBlobImages() {
        document.querySelectorAll('a[href*=".blob"]').forEach(el => {
            const img = document.createElement('img');
            img.className = 'blob-image';
            img.loading = 'lazy';
            img.src = el.href.replace('.blob', '.jpg');
            el.parentElement.replaceChild(img, el);
        });
    }

    renderBlobImages();
});

while this does the job, I'd prefer handling this in Python where it belongs.

This pull request will no longer be automatically closed when a new version is found as this pull request was created by Dependabot Preview and this repo is using a version: 2 config file. You can close this pull request and let Dependabot re-create it the next time it checks for updates.

Superseded by #1313.

Superseded by #1311.

@dracos I appreciate your ideas!

1. Ooh, I like this: https://codepen.io/astro87/pen/LYRQNbd?editors=1100 (That's the codepen from your linked stackoverflow.)
2. I worry that a max height will be a problem when my facets are open. (I've got 35 active ingredients, and so I've set the default_facet_size to 35.)
3. I don't understand this one. I'm observing the screenshot... very helpful! (Ah, okay, TR = Top Right and BR = Bottom Right. Absolute grid refers to position style.) All the scroll bars look a little wonky to me. I've also got a lot of facets, and prefer the extra horizontal space so that not as many facets disappear below the fold. My site also has end users... some will be on mobile... not sure what the absolute grid would do there...
4. (I still think a hover-arrow that scrolls upon click would help, too...)

But meanwhile, I'm going to go ahead and see if I can apply that shadow. (Never would've thought of that.) Hmmm... I'm not an SCSS person. This looks helpful! https://jsonformatter.org/scss-to-css

If there's a simpler way to generate a URL for a specific row, I'm all ears

If I change the hookspec and add a row parameter, it works

https://github.com/simonw/datasette/blob/7a2ed9f8a119e220b66d67c7b9e07cbab47b1196/datasette/hookspecs.py#L58

def render_cell(value, column, row, table, database, datasette):

But to generate a URL, I need the primary keys, but I can't call pks = await db.primary_keys(table) inside a sync function. I can't call datasette.utils.detect_primary_keys either, because the db connection is not publicly exposed (AFAICT).

Removing /var/lib/apt and /var/lib/dpkg makes apt and dpkg unusable in
images based on this one. Running apt-get clean and removing
/var/lib/apt/lists achieves similar size savings.

this PR helps me as removing the /var/lib/apt and /var/lib/dpkg directories breaks my ability to add packages when using datasetteproject/datasette:0.56 as a base image.

Shorterm workaround for me was to use this in my Dockerfile

FROM datasetteproject/datasette:0.56

RUN mkdir -p /var/lib/apt
RUN mkdir -p /var/lib/dpkg
RUN mkdir -p /var/lib/dpkg/updates
RUN mkdir -p /var/lib/dpkg/info
RUN touch /var/lib/dpkg/status

RUN apt-get update # and install your packages etc

Related: #1285, including arguments for natural breaks, equal interval, etc. modeled after choropleth map legends.

This worked for me:
<td class="col-{{ cell.column|to_css_class }} type-{{ cell.value_type }}">{{ cell.value | replace('", "','; ') | replace('[\"','') | replace('\"]','')}}</td>

I'm sure there is a prettier (and more flexible) way, but for now, this is ever-so-much more pleasant to look at.

------ AFTER:
https://user-images.githubusercontent.com/192568/114062829-f2f20c00-9865-11eb-891d-b0c348a9b433.png">

------ BEFORE:
https://user-images.githubusercontent.com/192568/114062871-01402800-9866-11eb-91ce-91efe4ee45cd.png">

(Note: I didn't figure out how to have one item have no semicolon, while multi-items close with a semicolon, but this is good enough for now. I also didn't figure out how to set up a new jinja filter. I don't want to add to /datasette/utils/init.py as I assume that would get overwritten when upgrading datasette. Having a starter guide on creating jinja filters in datasette would be helpful. (The jinja documentation isn't datasette-specific enough for me to quite nail it.)

For what it's worth, if anyone finds this in the future, I was having the same issue.

After digging through the code, it turned out that the database download is only available if it the db served in immutable mode, so datasette serve -i xyz.db rather than the doc's quickstart recommendation of datasette serve xyz.db.

This shows the city in a different color (and not the comma), but I get the idea, and I like it. (Ooh, could be nice to have the gear have an option in array fields to show as bullets or commas or semicolons...)

Maybe the addition of two template files: 'one_database_index.html' and 'one_table_index.html' would be a better idea than the documentation diff idea. (They could include commented instructions to rename the preferred template 'index.html', along with any other necessary guidance.)

One issue I could see with adding first class support for metadata in hjson format is that this would require adding an additional dependency to handle this, for a feature that would be unused by many users. I wonder if this could fit in as a plugin instead; if a hook existed for loading metadata (maybe as part of https://github.com/simonw/datasette/issues/860) the metadata could then come from any source, as specified by plugins, e.g. hjson, toml, XML, a database table etc.

Until/unless this exists, a few ideas for how you could add comments:
- Using YAML as you suggest.
- A common pattern is adding a "comment" key for comments to any object in JSON - I don't think including an unnecessary key like this would break anything in Datasette, but not certain.
- You could use another tool as a preprocessor for your JSON metadata - e.g. hjson or Jsonnet. You'd write the metadata in that format, and then convert that into JSON to actually use as your final metadata.

Could there be a little widget that offers conversion from one to the other?

With just three facets, I like it, but it does take more horizontal space. Would be nice to have a switch somewhere, enabling either original compact option or this proposed more-readable option. Also some control over word wrap (width setting) and facet spacing.

"datasette skeleton" feature removed #476

I'm visiting old issues for context while learning datasette. Let me know if okay to make the occasional comment like this one.
querystring argument now located at:
https://docs.datasette.io/en/latest/settings.html#sql-time-limit-ms

Google Map API limits seem to prevent https://nhs-england-map.netlify.com from being a working demo.

Sounds like a good idea.

I like this idea. Though it might be nice to have some kind of automated system from database to file, so that developers could easily track diffs.

It doesn't use basic auth, but you can put a whole datasette instance, or parts of this, behind a username/password prompt using https://github.com/simonw/datasette-auth-passwords

This sounds like a good use case for a plugin, since this will only be useful for a subset of Datasette users. It shouldn't be too difficult to add a button to do this with the available plugin hooks - have you taken a look at https://docs.datasette.io/en/latest/writing_plugins.html?

Oh, and the container image can be up to 10GB, so the EFS step might not be needed except for pretty big stuff.

Now that Lambda supports Docker, this probably is a bit easier and may be able to build on top of the existing package command.

There are weirdnesses in how the command actually gets invoked; the aws-lambda-python image shows a bit of that. So Datasette would probably need some sort of Lambda-specific entry point to make this work.

Sorry, I meant "minor typo" not "minor type".

Hello,
I have the same wildcards search problems with an instance of Datasette. http://crbc-dataset.huma-num.fr/inventaires/fonds_auguste_dupouy_1872_1967?_search=gwerz&_sort=rowid is OK but http://crbc-dataset.huma-num.fr/inventaires/fonds_auguste_dupouy_1872_1967?_search=gwe* is not (FTS is activated on "Reference" "IntituleAnalyse" "NomDuProducteur" "PresentationDuContenu" "Notes").

Notice that a SQL query as below launched directly from SQLite in the server's shell, retrieves results.

select * from fonds_auguste_dupouy_1872_1967_fts where IntituleAnalyse MATCH "gwe*";

Thanks,

A custom templates/index.html seems to work and custom pages as a workaround with moving them to pages/base_url_dir.

I will close this issue since it appears only in my particular setup.

I believe this pr and #1031 are related and fix the same issue.

That Docker command will mount your current directory inside the Docker container at /mnt - so you shouldn't need to change anything locally, just run

docker run -p 8001:8001 -v `pwd`:/mnt \
    datasetteproject/datasette \
    datasette -p 8001 -h 0.0.0.0 /mnt/fixtures.db

and it will use the fixtures.db file within your current directory

For the purposes below I am assuming the organization I would get all the repositories and their related commits from is called gh-organization. The github's owner id of gh-orgnization is 123456789.

github-to-sqlite  repos github.db gh-organization

I'm on a windows computer running git bash to be able to use the | command. This works for me

sqlite3 github.db "SELECT full_name FROM repos WHERE owner = '123456789';" | tr '\n\r' ' ' | xargs | { read repos; github-to-sqlite commits github.db $repos; }

On a pure linux system I think this would work because the new line character is normally \n

sqlite3 github.db "SELECT full_name FROM repos WHERE owner = '123456789';" | tr '\n' ' ' | xargs | { read repos; github-to-sqlite commits github.db $repos; }`

As expected I ran into rate limit issues #51

Yes, Big Sur Photos database doesn't have ZGENERICASSET table. PR #31 will fix this.

According to https://github.com/simonw/datasette/blob/master/docs/installation.rst#using-docker it should be

docker run -p 8001:8001 -v `pwd`:/mnt \
    datasetteproject/datasette \
    datasette -p 8001 -h 0.0.0.0 /mnt/fixtures.db

This uses /mnt/fixtures.db whereas you're using fixtures.db - did you try using this path instead?

I'm probably missing something but what's the use case here - what would this offer over adding limit 10 to the query?

@simonw could you please take a look at the PR 1211 that fixes this issue?

How do you get 4-5 minutes?
I run my tests in WSL 2, so may be i need to try a real linux VM.

Ping @simonw

I don't have much experience with github's rate limiting. In my day job we use the tenacity library to handle http errors we get.

Yes that would be cool! I wouldn't mind helping. Is this the meat of it? https://github.com/dogsheep/twitter-to-sqlite/blob/21fc1cad6dd6348c67acff90a785b458d3a81275/twitter_to_sqlite/utils.py#L512

It looks like the cli option is added with this decorator : https://github.com/dogsheep/twitter-to-sqlite/blob/21fc1cad6dd6348c67acff90a785b458d3a81275/twitter_to_sqlite/cli.py#L14

I looked a bit at utils.py in the GitHub repository. I was surprised at the amount of manual mapping of the API response you had to do to get this to work.

Not sure where exactly to put the actual docs (presumably somewhere in docs/contributing.rst) but I've made a slight change to make it easier to run locally (copying the approach in excalidraw): https://github.com/simonw/datasette/compare/main...benpickles:prettier-docs

I'm seeing the problem on database page. Index page and table page runs quite fast.

• Tables have <10 columns (id, url, title, body_html, date, author, meta (for keeping unstructured json)). I've added index on date columns (using sqlite-utils) in addition to the index present on id columns.
• All tables have FTS enabled on text and varchar columns (title, body_html etc) to speed up searching.
• There are couple of tables related with foreign keys (think a thread in a forum and posts in that thread, related with thread_id)

I somewhat share your reluctance to add a package.json to seemingly every project out there but ultimately if they're project dependencies it's important they're managed within the codebase.

I was going to re-add the filter to only run Prettier when there have been changes in datasette/static but that would mean it wouldn't run when the package is updated. That plus the fact that the last run of the job took only 8 seconds is why I decided not to re-add the filter.

Actually, on more research, I found out this is handled by the trove-classifiers package now, so it's just a one-liner pr instead of fire-up-a-docker-container-and-do-some-migrations

FWIW, I had a look at watchdog for a datasette powered Jupyter notebook search tool: https://github.com/ouseful-testing/nbsearch/blob/main/nbsearch/nbwatchdog.py

Not a production thing, just an experiment trying to explore what might be possible...

Hi @simonw, do I need to do anything else for this PR to be considered to be included? I've tried using this project and it is quite nice to be able to explore a repository, but noticed that a couple commands don't allow you to use authorization from the environment variable.

The dict/set suggestion comes from pyupgrade --py36-plus, but then had to black the change.

The rest comes from PyCharm's Inspect code function. I reviewed all the suggestions and fixed a thing or two, such as leading/trailing spaces in the docstrings or turned around the chained conditions.

Then I tried to convert all os.path/glob/open to Path, but there were some local test issues, so I'll have to start over in smaller chunks if you want to have that too.

Fixes the issue mentioned here: https://github.com/dogsheep/dogsheep-photos/issues/15#issuecomment-748436115

@nickvazz @simonw I opened a PR that replaces the SQL for ZCOMPUTEDASSETATTRIBUTES to use osxphotos which now exposes all this data and has been updated for Big Sur. I did regression tests to confirm the extracted data is identical, with one exception which should not affect operation: the old code pulled data from ZCOMPUTEDASSETATTRIBUTES for missing photos while the main loop ignores missing photos and does not add them to apple_photos. The new code does not add rows to the apple_photos_scores table for missing photos.

@nickvazz ZGENERICASSET changed to ZASSET in Big Sur. Here's a list of other changes to the schema in Big Sur: https://github.com/RhetTbull/osxphotos/wiki/Changes-in-Photos-6---Big-Sur

I can't keep up with the renaming contexts, but I like having the ability to run datasette+ datasette-ripgrep against different configs:

datasette serve --metadata=./metadata.json

I have one for all of my code and one per client who has lots of code. So as long as I can point to datasette to something, it's easy to work with.

Hello Simon,
I have a similar problem with horizontal scrollbar display with Datasette version 0.51 and superior for a table with more than 30 rows. With Datasette 0.50, the horizontal scrollbar is displayed, if I upgrade Datasette to 0.51 and superior, the horizontal scrollbar disappears.

Datasette 0.50: horizontal scrollbar

Datasette 0.51 and superior: no horizontal scrollbar

Thanks,

EDIT: I misunderstood the problem. This seems like a fix better suited for Safari. But I don't have any Apple device to test it.

body {
  min-height: 100vh;
  min-height: -webkit-fill-available;
}
html {
  height: -webkit-fill-available;
}

https://css-tricks.com/css-fix-for-100vh-in-mobile-webkit/

It's actually not that difficult to fix.
Well, this is actually a workaround to keep viewport in place.

I usually put a transition (forgot to do it here) that keeps page from resizing.

.container {
  min-height: 100vh;
  transition: height 10000s steps(0);
}

steps() function prevents excessive layout calculations, and lets the page snap back into place (10000s ~= 3h later) in a single step.
This fix also prevents page from jumping around when the keyboard pops up and down.

Somewhat related: https://github.com/simonw/datasette/issues/859
I fixed the issue with forking and disabling the counts for hidden tables.

I was confused by --config and I tried passing the json from datasette-ripgrep into config.json just as a wild guess.

A short term solution might be pointing out in plugins that their snippet json can go in metadata.json at least makes it easier to search for config options or to know where to start if someone is new.

(Even more annoying is that I see my editor leaked an extra delete space at the end of the line. I'm happy to rebuild this to be less annoying, but you probably don't want the changelog update either way)

Thanks; just a note that the datasette.urls.static(path) and datasette.urls.static_plugins(plugin_name, path) items both seem to be repeated and appear in the docs twice?

That workaround is probably fine. I was trying to work out whether there might be other situations where a pre-external package load might be useful but couldn't offhand bring any other examples to mind. The static plugins option also looks interesting.

No problem. I added a test. Let me know if it looks sufficient or if you want me to to tweak something!

If you don't mind, would you tag this PR as "hacktoberfest-accepted"? If you do mind, no problem and I'm sorry for asking :) My kiddos like the shirts.

Sorry, I was out of the loop this weekend. The missing sdists were in some the datasette-* plugins... i'll capture my findings more concretely in one spot when i have a chance...

I'm trying to run something behind a MyBinder proxy, but seem to have something set up incorrectly and not sure what the fix is?

I'm starting datasette with jupyter-server-proxy setup:

# __init__.py
def setup_nbsearch():

    return {
        "command": [
            "datasette",
            "serve",
            f"{_NBSEARCH_DB_PATH}",
            "-p",
            "{port}",
            "--config",
            "base_url:{base_url}nbsearch/"
        ],
        "absolute_url": True,
        # The following needs a the labextension installing.
        # eg in postBuild: jupyter labextension install jupyterlab-server-proxy
        "launcher_entry": {
            "enabled": True,
            "title": "nbsearch",
        },
    }

where the base_url gets automatically populated by the server-proxy. I define the loaders as:

# __init__.py
from datasette import hookimpl

@hookimpl
def extra_css_urls(database, table, columns, view_name, datasette):
    return [
        "/-/static-plugins/nbsearch/prism.css",
        "/-/static-plugins/nbsearch/nbsearch.css",
    ]

but these seem to also need a base_url prefix set somehow?

Currently, the generated HTML loads properly but internal links are incorrect; eg they take the form <link rel="stylesheet" href="/-/static-plugins/nbsearch/prism.css"> which resolves to eg https://notebooks.gesis.org/hub/-/static-plugins/nbsearch/prism.css rather than required URL of form https://notebooks.gesis.org/binder/jupyter/user/ouseful-testing-nbsearch-0fx1mx67/nbsearch/-/static-plugins/nbsearch/prism.css.

The main css is loaded correctly: <link rel="stylesheet" href="/binder/jupyter/user/ouseful-testing-nbsearch-0fx1mx67/nbsearch/-/static/app.css?404439">

From the docs, I note:

datasette.urls.instance()
Returns the URL to the Datasette instance root page. This is usually "/"

What about the proxy case? Eg if I am using jupyter-server-proxy on a MyBinder or local Jupyter notebook server site, https://example.com:PORT/weirdpath/datasette, what does datasette.urls.instance() refer to?

• https://example.com:PORT/weirdpath/datasette
• https://example.com:PORT/weirdpath/
• https://example.com:PORT/
• https://example.com
• something else?

Good luck on 1.0! It may also be worth lobbying for a Framework::Datasette::1.0 classifier. This would be a nice way to allow the ecosystem to self-document a bit more discoverably.

I was surprised to see the PR for Framework::Jupyter is a... database migration! Of course, there may be more workflow to it!

How does /-/static relate to current guidance docs around static regarding the --static option and metadata formulations such as "extra_js_urls": [ "/static/app.js"] (I've not managed to get this to work in a Jupyter server proxied set up; the datasette / jupyter server proxy repo may provide a useful test example, eg via MyBinder, for folk to crib from?)

🎉 Thanks so much @simonw ! 🎉

This closes issue #8

ref #46

Okay, I've rewritten this PR to preserve the batching behaviour but still fix #145, and rebased the branch to account for the db.execute() api change. It's not terribly sophisticated -- if it attempts to insert a batch which has too many variables, the exception is caught, the batch is split in two and each half is inserted separately, and then it carries on as before with the same batch_size. In the edge case where this gets triggered, subsequent batches will all be inserted in two groups too if they continue to have the same number of columns (which is presumably reasonably likely). Do you reckon this is acceptable when set against the awkwardness of recalculating the batch_size on the fly?

Just force-pushed to update d042f9c with more formatting changes to satisfy black==20.8b1 and pass the GitHub Actions "Test" workflow.

@simonw -- I've gone ahead updated the documentation to reflect the changes introduced in this PR. IMO it's ready to merge now.

In writing the documentation changes, I begin to wonder about the value and role of batch_size at all, tbh. May I assume it was originally intended to prevent using the entire row set to determine columns and column types, and that this was a performance consideration? If so, this PR entirely undermines its purpose. I've been passing in excess of 500,000 rows at a time to insert_all() with these changes and although I'm sure the performance difference is measurable it's not really noticeable; given #145, I don't know that any performance advantages outweigh the problems doing it this way removes. What do you think about just dropping the argument and defaulting to the maximum batch_size permissible given SQLITE_MAX_VARS? Are there other reasons one might want to restrict batch_size that I've overlooked? I could open a new issue to discuss/implement this.

Of course the documentation will need to change again too if/when something is done about #147.

Looks like black is up-to-date now, so this is no longer needed.

Note: had to adjust the test above because trying to exhaust a SQLITE_MAX_VARIABLE_NUMBER of 250000 in 99 records requires 2526 columns, and trips the "Rows can have a maximum of {} columns".format(SQLITE_MAX_VARS) check even before it trips the default SQLITE_MAX_COLUMN value (2000).

Thanks! And yeah, I had updating the docs on my list too :) Will try to get to it this afternoon (budgeting time is fraught with uncertainty at the moment!).

I tried changing the batch_size argument to the total number of records, but it seems only to effect the number of rows that are committed at a time, and has no influence on this problem.

So the reason for this is that the batch_size for import is limited (of necessity) here: https://github.com/simonw/sqlite-utils/blob/main/sqlite_utils/db.py#L1048

With regard to the issue of ignoring columns, however, I made a fork and hacked a temporary fix that looks like this:
https://github.com/simonwiles/sqlite-utils/commit/3901f43c6a712a1a3efc340b5b8d8fd0cbe8ee63

It doesn't seem to affect performance enormously (but I've not tested it thoroughly), and it now does what I need (and would expect, tbh), but it now fails the test here:
https://github.com/simonw/sqlite-utils/blob/main/tests/test_create.py#L710-L716

The existence of this test suggests that insert_all() is behaving as intended, of course. It seems odd to me that this would be a desirable default behaviour (let alone the only behaviour), and its not very prominently flagged-up, either.

@simonw is this something you'd be willing to look at a PR for? I assume you wouldn't want to change the default behaviour at this point, but perhaps an option could be provided, or at least a bit more of a warning in the docs. Are there oversights in the implementation that I've made?

Would be grateful for your thoughts! Thanks!

Bumping this, as the previous fix is passing the wrong type, and not actually addressing the issue...

The exclude argument needs an iterable of packages instead of a single string (but since str is iterable, it's currently excluding packages t, e, and s.)

Yep, I agree that makes more sense for backwards compat and more casual use cases. I think it should be possible for the Database/Queryable methods to DTRT based on seeing if it's within a context-manager-managed transaction.

Better transaction handling would be really great. Some of my thoughts on implementing better transaction discipline are in https://github.com/simonw/sqlite-utils/pull/118#issuecomment-655239728.

My preferences:

• Each CLI command should operate in a single transaction so that either the whole thing succeeds or the whole thing is rolled back. This avoids partially completed operations when an error occurs part way through processing. Partially completed operations are typically much harder to recovery from gracefully and may cause inconsistent data states.

• The Python API should be transaction-agnostic and rely on the caller to coordinate transactions. Only the caller knows how individual insert, create, update, etc operations/methods should be bundled conceptually into transactions. When the caller is the CLI, for example, that bundling would be at the CLI command-level. Other callers might want to break up operations into multiple transactions. Transactions are usually most useful when controlled at the application-level (like logging configuration) instead of the library level. The library needs to provide an API that's conducive to transaction use, though.

• The Python API should provide a context manager to provide consistent transactions handling with more useful defaults than Python's sqlite3 module. The latter issues implicit BEGIN statements by default for most DML (INSERT, UPDATE, DELETE, … but not SELECT, I believe), but not DDL (CREATE TABLE, DROP TABLE, CREATE VIEW, …). Notably, the sqlite3 module doesn't issue the implicit BEGIN until the first DML statement. It does not issue it when entering the with conn block, like other DBAPI2-compatible modules do. The with conn block for sqlite3 only arranges to commit or rollback an existing transaction when exiting. Including DDL and SELECTs in transactions is important for operation consistency, though. There are several existing bugs.python.org tickets about this and future changes are in the works, but sqlite-utils can provide its own API sooner. sqlite-utils's Database class could itself be a context manager (built on the sqlite3 connection context manager) which additionally issues an explicit BEGIN when entering. This would then let Python API callers do something like:

db = sqlite_utils.Database(path)

with db: # ← BEGIN issued here by Database.__enter__
    db.insert(…)
    db.create_view(…)
# ← COMMIT/ROLLBACK issue here by sqlite3.connection.__exit__

The only thing missing from this PR is updates to the documentation.

Ah, yes, thanks for this reminder! I've repushed with doc bits added.

I fixed my original oops by moving the DELETE FROM $table out of the chunking loop and repushed. I think this change can be considered in isolation from issues around transactions, which I discuss next.

I wanted to make the DELETE + INSERT happen all in the same transaction so it was robust, but that was more complicated than I expected. The transaction handling in the Database/Table classes isn't systematic, and this poses big hurdles to making Table.insert_all (or other operations) consistent and robust in the face of errors.

For example, I wanted to do this (whitespace ignored in diff, so indentation change not highlighted):

diff --git a/sqlite_utils/db.py b/sqlite_utils/db.py
index d6b9ecf..4107ceb 100644
--- a/sqlite_utils/db.py
+++ b/sqlite_utils/db.py
@@ -1028,6 +1028,11 @@ class Table(Queryable):
         batch_size = max(1, min(batch_size, SQLITE_MAX_VARS // num_columns))
         self.last_rowid = None
         self.last_pk = None
+        with self.db.conn:
+            # Explicit BEGIN is necessary because Python's sqlite3 doesn't
+            # issue implicit BEGINs for DDL, only DML.  We mix DDL and DML
+            # below and might execute DDL first, e.g. for table creation.
+            self.db.conn.execute("BEGIN")
             if truncate and self.exists():
                 self.db.conn.execute("DELETE FROM [{}];".format(self.name))
             for chunk in chunks(itertools.chain([first_record], records), batch_size):
@@ -1038,7 +1043,11 @@ class Table(Queryable):
                         # Use the first batch to derive the table names
                         column_types = suggest_column_types(chunk)
                         column_types.update(columns or {})
-                    self.create(
+                        # Not self.create() because that is wrapped in its own
+                        # transaction and Python's sqlite3 doesn't support
+                        # nested transactions.
+                        self.db.create_table(
+                            self.name,
                             column_types,
                             pk,
                             foreign_keys,
@@ -1139,7 +1148,6 @@ class Table(Queryable):
                     flat_values = list(itertools.chain(*values))
                     queries_and_params = [(sql, flat_values)]

-            with self.db.conn:
                 for query, params in queries_and_params:
                     try:
                         result = self.db.conn.execute(query, params)

but that fails in tests because other methods call insert/upsert/insert_all/upsert_all in the middle of their transactions, so the BEGIN statement throws an error (no nested transactions allowed).

Stepping back, it would be nice to make the transaction handling systematic and predictable. One way to do this is to make the sqlite_utils/db.py code generally not begin or commit any transactions, and require the caller to do that instead. This lets the caller mix and match the Python API calls into transactions as appropriate (which is impossible for the API methods themselves to fully determine). Then, make sqlite_utils/cli.py begin and commit a transaction in each @cli.command function, making each command robust and consistent in the face of errors. The big change here, and why I didn't just submit a patch, is that it dramatically changes the Python API to require callers to begin a transaction rather than just immediately calling methods.

There is also the caveat that for each transaction, an explicit BEGIN is also necessary so that DDL as well as DML (as well as SELECTs) are consistent and rolled back on error. There are several bugs.python.org discussions around this particular problem of DDL and some plans to make it better and consistent with DBAPI2, eventually. In the meantime, the sqlite-utils Database class could be a context manager which supports the incantations necessary to do proper transactions. This would still be a Python API change for callers but wouldn't expose them to the weirdness of the sqlite3's default transaction handling.

Ah, I see the problem. The truncate is inside a loop I didn't realize was there.

Hmm, while tests pass, this may not work as intended on larger datasets. Looking into it.