Summary:
If the longer OpenAI model exceeds the OpenAPI context length, we now perform another retry with a
shorter variant of the formula prompt. The shorter prompt excludes non-referenced tables and lookup
method definitions, which should help reduce token usage in documents with larger schemas.
Test Plan: Server test.
Reviewers: JakubSerafin
Reviewed By: JakubSerafin
Subscribers: JakubSerafin
Differential Revision: https://phab.getgrist.com/D4184
Summary:
It was manifesting as error "docactions.[Bulk]UpdateRecord for
non-existent record #1", and due to manualSort column having internal
state that wasn't properly cleared.
Test Plan: Added a test case that fails without the fix.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Subscribers: georgegevoian
Differential Revision: https://phab.getgrist.com/D4183
Summary:
Improvements
- Widget and column descriptions are now copied when duplicating a table.
- A Grist Plugin API command to open a Record Card is now available.
- New Card widgets set initial settings based on those used by their table's
Record Card.
Fixes
- Opening a reference in a Record Card from a Raw Data popup now opens
the correct reference.
Test Plan: Browser and python tests.
Reviewers: jarek
Reviewed By: jarek
Differential Revision: https://phab.getgrist.com/D4164
* run.sh: Replace pid with nodejs's one using exec
This notably ensures that "docker stop" sends a TERM signal to the node
process which can handle it gracefully and avoid document corruption.
* Use exec form for CMD in Dockerfile
---------
Co-authored-by: Florent FAYOLLE <florent.fayolle@beta.gouv.fr>
Summary:
Forms improvements and following new design
- New headers
- New UI
- New right panel options
Test Plan: Tests updated
Reviewers: georgegevoian, dsagal
Reviewed By: georgegevoian
Subscribers: dsagal, paulfitz
Differential Revision: https://phab.getgrist.com/D4158
Summary: Adds `@functools.lru_cache` to `ts_to_dt` and `ts_to_date` which are used in `_make_rich_value` in Date/DateTime columns, i.e. every time such a column is accessed in a formula. I noticed that these operations are surprisingly expensive while working on https://phab.getgrist.com/D4075. This is just an easy way to potentially significantly speed up certain docs and formulas.
Test Plan:
Put this code in an engine test case:
```
def test_(self):
self.apply_user_action(["AddTable", "Table1", [
{"id": "A", "type": "DateTime:America/Chicago"},
]])
self.add_records("Table1", ["A"], [
[i] for i in range(1000)
])
formula = "for _ in range(1000): $A\nreturn 1"
import time
start = time.time()
self.add_column(
"Table1", formula, type="Any", isFormula=True, formula=formula
)
elapsed = time.time() - start
print(f"Took {elapsed:.2f} for formula {formula}")
```
The time goes from ~10s to ~1s. Similar tests showed no noticeable slowdown when caching had no effect.
Reviewers: dsagal
Reviewed By: dsagal
Differential Revision: https://phab.getgrist.com/D4099
Summary:
This gives a mechanism for controlling access control within a document that is distinct from (though implemented with the same machinery as) granular access rules.
It was hard to find a good way to insert this that didn't dissolve in a soup of complications, so here's what I went with:
* When reading rules, if there are shares, extra rules are added.
* If there are shares, all rules are made conditional on a "ShareRef" user property.
* "ShareRef" is null when a doc is accessed in normal way, and the row id of a share when accessed via a share.
There's no UI for controlling shares (George is working on it for forms), but you can do it by editing a `_grist_Shares` table in a document. Suppose you make a fresh document with a single page/table/widget, then to create an empty share you can do:
```
gristDocPageModel.gristDoc.get().docData.sendAction(['AddRecord', '_grist_Shares', null, {linkId: 'xyz', options: '{"publish": true}'}])
```
If you look at the home db now there should be something in the `shares` table:
```
$ sqlite3 -table landing.db "select * from shares"
+----+------------------------+------------------------+--------------+---------+
| id | key | doc_id | link_id | options |
+----+------------------------+------------------------+--------------+---------+
| 1 | gSL4g38PsyautLHnjmXh2K | 4qYuace1xP2CTcPunFdtan | xyz | ... |
+----+------------------------+------------------------+--------------+---------+
```
If you take the key from that (gSL4g38PsyautLHnjmXh2K in this case) and replace the document's urlId in its URL with `s.<key>` (in this case `s.gSL4g38PsyautLHnjmXh2K` then you can use the regular document landing page (it will be quite blank initially) or API endpoint via the share.
E.g. for me `http://localhost:8080/o/docs/s0gSL4g38PsyautLHnjmXh2K/share-inter-3` accesses the doc.
To actually share some material - useful commands:
```
gristDocPageModel.gristDoc.get().docData.getMetaTable('_grist_Views_section').getRecords()
gristDocPageModel.gristDoc.get().docData.sendAction(['UpdateRecord', '_grist_Views_section', 1, {shareOptions: '{"publish": true, "form": true}'}])
gristDocPageModel.gristDoc.get().docData.getMetaTable('_grist_Pages').getRecords()
gristDocPageModel.gristDoc.get().docData.sendAction(['UpdateRecord', '_grist_Pages', 1, {shareRef: 1}])
```
For a share to be effective, at least one page needs to have its shareRef set to the rowId of the share, and at least one widget on one of those pages needs to have its shareOptions set to {"publish": "true", "form": "true"} (meaning turn on sharing, and include form sharing), and the share itself needs {"publish": true} on its options.
I think special shares are kind of incompatible with public sharing, since by their nature (allowing access to all endpoints) they easily expose the docId, and changing that would be hard.
Test Plan: tests added
Reviewers: dsagal, georgegevoian
Reviewed By: dsagal, georgegevoian
Subscribers: jarek, dsagal
Differential Revision: https://phab.getgrist.com/D4144
Summary: Descriptions can now be set on Raw Data table sections.
Test Plan: Browser tests.
Reviewers: jarek
Reviewed By: jarek
Differential Revision: https://phab.getgrist.com/D4131
Summary: Title
Test Plan: Fixed a test that was failing inconsistently
Reviewers: georgegevoian, paulfitz
Reviewed By: georgegevoian, paulfitz
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D4122
Summary:
Adds a new Record Card view section to each non-summary table, which can be from opened from various parts of the Grist UI to view and edit records in a popup card view.
Work is still ongoing, so the feature is locked away behind a flag; follow-up work is planned to finish up the implementation and add end-to-end tests.
Test Plan: Python and server tests. Browser tests will be included in a follow-up.
Reviewers: jarek, paulfitz
Reviewed By: jarek
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D4114
Summary: Call a new user action `RemoveTransformColumns` in ActiveDoc shutdown.
Test Plan: Added nbrowser test
Reviewers: georgegevoian, paulfitz
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D4107
Pyodide packages needed rebuilding, and the pyodide project has
moved on a bit so a new version number is needed. The new packages
have already been built and pushed to S3.
To verify, go to `sandbox/pyodide` and follow the README there.
Then at top level do `GRIST_SANDBOX_FLAVOR=pyodide`. Try to
create and edit a document using formulas. It should work.
Summary: While looking at webhooks code, I noticed that it calls `ActiveDoc.fetchQuery` (which I think typically leads to the sandbox method `fetch_table`) with a query on the `id` column, where the values could potentially be all row IDs in the table (e.g. if a new column was added, as in a recent discussion about webhooks gone wrong). This suggests that `fetch_table` should try to avoid using a list for values when possible. In practice this only starts to become an issue at about 10k rows, so I don't know if this has caused any real problems, but it seemed like something worth fixing.
Test Plan:
Extended unit tests for correctness. Tested performance manually. Made a doc with this formula:
```
import time
row_ids = list(Table2.table.row_ids)
query = {"id": row_ids}
start = time.time()
result = table.table._engine.fetch_table('Table2', query=query)
end = time.time()
assert result[1] == row_ids
end - start, len(row_ids)
```
Then put a bunch of rows in `Table2`. This diff made the returned elapsed time much less.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Subscribers: jarek
Differential Revision: https://phab.getgrist.com/D4092
Summary:
Fixes a very specific bug reported here: https://grist.slack.com/archives/C069RUP71/p1694630242765769
The error occurred when:
1. Removing multiple columns simultaneously
2. Those columns were sources of groupby columns for a summary table (so removing them meant recreating the summary table and thus deleting its columns)
3. There was a display column for one of the columns that got deleted (either directly or indirectly) which was set to be automatically removed since it was no longer needed, but this failed because the column was already deleted as part of earlier table removal.
I fixed this by making `apply_auto_removes` remove table records last, so removing the display column wouldn't be a problem.
That fixed the original error, but then I noticed that trying to undo the removal could lead to another error (under even more specific circumstances). It's hard to see exactly why, but I can see that just 3 `RemoveColumn` user actions generated over 100 doc actions and corresponding undo actions, hence the difficulty in narrowing the problem down. This is partly because removing a single column would recreate a summary table, only for that table to be immediately replaced again when another column was removed. Making the frontend send a single `[BulkRemoveRecord, _grist_Tables_column, ...] ` leads to a more efficient and sensible process with about half as many doc actions and no undo error. I think this alone would also solve the original error, but the data engine change seems more generally helpful and worth keeping.
Test Plan: Added a Python test and an nbrowser test
Reviewers: jarek
Reviewed By: jarek
Differential Revision: https://phab.getgrist.com/D4052
Test Plan: Added a test case that reproduces the bug and tests the fix
Reviewers: alexmojaki
Reviewed By: alexmojaki
Subscribers: alexmojaki
Differential Revision: https://phab.getgrist.com/D4057
Summary: Upgrades asttokens and uses the newly released support for astroid trees in `asttokens.ASTText` which is needed to deal with f-strings (as opposed to `asttokens.ASTTokens`).
Test Plan: Added a Python unit test
Reviewers: dsagal
Reviewed By: dsagal
Differential Revision: https://phab.getgrist.com/D4027
Summary:
Previously, when a table was deleted, ref/reflist columns pointing at that table were deleted as well. This diff changes that, converting the columns to a suitable type instead.
See here for discussion and decisions: https://grist.slack.com/archives/C069RUP71/p1686060034848139
Test Plan: Added Python tests for most cases, and a DocApi test for where Python has to call JS code to convert the values.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D4011
A chardet dependency had changed, breaking a grist-electron upgrade.
So I reran the pyodide script to build the new one, and pushed it
to our CDN. This file changed. I don't see exactly where it plays
a role, but I thought I'd better commit it anyway.
Summary:
- Using a sample of data was causing poor detection if the sample were
cut mid-character. Switch to using line-based detection.
- Add a simple option for changing encoding. No convenient UI is offered
since config UI is auto-generated, but this at least makes it possible to
recover from bad guesses.
- Upgrades chardet library for good measure.
- Also fixes python3-building step, to more reliably rebuild Python
dependencies when requirements3.* files change.
Test Plan:
Added a python-side test case, and a browser test that encodings can
be switched, errors are displayed, and wrong encodings fail recoverably.
Reviewers: alexmojaki
Reviewed By: alexmojaki
Differential Revision: https://phab.getgrist.com/D3979
Summary:
Replaced uses of asttokens.ASTTokens with asttokens.ASTText when working with plain `ast` trees, and use `atok.get_text_range` instead of `node.first_token`.
Upgraded asttokens in Python 2 (it was already upgraded in Python 3).
Test Plan: Added a test with f-strings.
Reviewers: dsagal
Reviewed By: dsagal
Differential Revision: https://phab.getgrist.com/D4001
Summary:
Sometimes the model repeats the classes given in the prompt which would mess up extracting the actual formula. This diff solves this by:
1. Changes the generated Python schema so that (a) the thing that needs completing is a plain top level function instead of a property/method inside the class and (2) the classes are fully valid syntax, which makes it easier to
2. Remove classes from the parsed Python code when converting the completion to a formula.
3. Tweak the prompt wording to discourage including classes in general, especially because sometimes the model tries to solve the problem by defining extra methods/attributes/classes.
While I was at it, I changed type hints to use builtins (e.g. `list` instead of `List`) to prevent `from typing import List` which was happening sometimes and would look weird in a formula. Similarly I removed `@dataclass` since that also implies an import, and this also fits with the tweaked wording that the classes are fake.
Test Plan:
Added a new test case to the formula dataset which triggers the unwanted behaviour. The factors that seem to trigger the problem are (1) a small schema so the classes are easier to repeat and (2) the need to import modules, which the model wants to place before all other code. The case failed before this diff and succeeded after. The tweaked wording reduces the chances of repeating the classes but didn't eliminate it, so forcibly removing the classes in Python was needed.
There were also a couple of other existing cases where repeating the classes was observed before but not after.
Overall the score increased from 49 to 51 out of 69 (including the new case). At one point the score was 53, but changes in whitespace were enough to make it drop again.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D4000
* REQUEST now supports POST
* Add extra flag for enabling REQUEST, also update README and comments
Co-authored-by: John Cant <a.jonncant@gmail.com>
Co-authored-by: Alex Hall <alex.mojaki@gmail.com>
Summary: Mostly just changes to tests to make them more flexible.
Test Plan: Python tests pass locally with 3.10 and 3.11. Making tests run in CI on these versions will happen in grist-core.
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D3978
Summary:
Upgrades asttokens mainly because I suspected it would have helped with an error in a user's document (see https://grist.slack.com/archives/C0234CPPXPA/p1686145370484509) but we were unable to confirm that. Also adds a related test that I (wrongly) expected to have a similar problem before upgrading asttokens.
Upgrades wrapt for Python 3.11 support. Closes https://github.com/gristlabs/grist-core/issues/534.
Not upgrading the dependencies for Python 2 because pip is giving me errors when I try to install to the `thirdparty` folder. Both these upgrades are motivated by issues specific to Python 3 so this doesn't seem worth pursuing immediately.
Test Plan: Existing tests.
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D3973
Summary:
Replaces https://phab.getgrist.com/D3940, particularly to avoid doing potentially unwanted things automatically.
Adds optional fields `evaluateCurrentFormula?: boolean; rowId?: number` to `FormulaAssistanceContext` (part of `AssistanceRequest`). When `evaluateCurrentFormula` is `true`, calls a new function `evaluate_formula` in the sandbox which computes the existing formula in the column (regardless of anything the AI may have suggested) and uses that to generate an additional system message which is added before the user's message. In theory this could be used in an interface where users ask why a formula doesn't work, including possibly a formula suggested by the AI. For now, it's only used in `runCompletion_impl.ts` for experimenting.
Also cleaned up a bit, removing `_chatMode` which is always `true` now, and uses of `regenerate` which is always `false`.
Test Plan: Updated `runCompletion_impl` to optionally use the new feature, in which case it now scores 51/68 instead of 49/68.
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3970
Summary:
The image Jenkins used up to now was from 2021. We updated buildtools/jenkins-ec2/Dockerfile since then but couldn't use new image; we can now that an issue with mounts in gvisor has been resolved.
The new image uses newer Chrome (111 vs 95); 111 is the latest version that works with the webdriver version we are using (after that one, copy-pasting tests fail).
In addition, this adds a more precise way to maintain python dependencies: by specifying top-level dependencies in the new core/sandbox/requirements3.in. An updated build step compiles that into requirements3.txt, and syncs venv to that using pip-sync.
This addresses the issue that previously, removing a module from dependencies would not have caused the build to remove it from sandbox_venv3.
Test Plan: Existing tests pass on the new image.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3952
Summary:
- Replace logger module by the standard module 'logging'.
- When a log message from the sandbox includes newlines (e.g. for tracebacks),
keep those lines together in the Node log message.
Previously each line was a different message, making it difficult to view
tracebacks, particularly in prod where each line becomes a separate message
object.
- Fix assorted lint errors.
Test Plan: Added a test for the log-line splitting and escaping logic.
Reviewers: georgegevoian
Reviewed By: georgegevoian
Differential Revision: https://phab.getgrist.com/D3956
Summary:
This tweaks the prompting so that the user's message is given on its own instead of as a docstring within Python. This is so that the prompt makes sense when:
- the user asks a question such as "Can you write me a formula which does ...?" rather than describing their formula as a docstring would, or
- the user sends a message that doesn't ask for a formula at all (https://grist.slack.com/archives/C0234CPPXPA/p1687699944315069?thread_ts=1687698078.832209&cid=C0234CPPXPA)
Also added wording for the model to refuse when the user asks for something that the model cannot do.
Because the code (and maybe in some cases the model) for non-ChatGPT models relies on the prompt consisting entirely of Python code produced by the data engine (which no longer contains the user's message) those code paths have been disabled for now. Updating them now seems like undesirable drag, I think it'd be better to revisit this when iteration/experimentation has slowed down and stabilised.
Test Plan:
Added entries to the formula dataset where the response shouldn't contain a formula, indicated by the value `1` for the new column `no_formula`.
This is somewhat successful, as the model does refuse to help in some of the new test cases, but not all. Performance on existing entries also seems a bit worse, but it's hard to distinguish this from random noise. Hopefully this can be remedied in the future with more work, e.g. automatic followup messages containing example inputs and outputs.
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: dsagal
Differential Revision: https://phab.getgrist.com/D3936
Summary: Converts `rec.` to `$` in AI generated formulas, and removes redundant `return` at the end.
Test Plan: Expanded unit test.
Reviewers: dsagal
Reviewed By: dsagal
Differential Revision: https://phab.getgrist.com/D3933
Summary:
The previous code for extracting a Python formula from the LLM completion involved some shaky string manipulation which this improves on.
Overall the 'test results' from `runCompletion` went from 37/47 to 45/47 for `gpt-3.5-turbo-0613`.
The biggest problem that motivated these changes was that it assumed that code was always inside a markdown code block
(i.e. triple backticks) and so if there was no block there was no code. But the completion often consists of *only* code
with no accompanying explanation or markdown. By parsing the completion in Python instead of JS,
we can easily check if the entire completion is valid Python syntax and accept it if it is.
I also noticed one failure resulting from the completion containing the full function (instead of just the body)
and necessary imports before that function instead of inside. The new parsing moves import inside.
Test Plan: Added a Python unit test
Reviewers: paulfitz
Reviewed By: paulfitz
Subscribers: paulfitz
Differential Revision: https://phab.getgrist.com/D3922
Summary:
In the future, it may be helpful to capture some output to display in the
formula editor (e.g. the result of calling print()), but for now it's best
to redirect stdout/stderr to avoid excessive logging.
Test Plan: Tested manually.
Reviewers: alexmojaki
Reviewed By: alexmojaki
Subscribers: alexmojaki
Differential Revision: https://phab.getgrist.com/D3919
Summary:
For grist-static, we want to the data engine to be able to call external/exported JS functions directly,
rather than via the node 'server' living in another thread which requires synchronous communication hackery.
As a step in that direction, this diff changes the exported functions that we care about (guessColInfo and convertFromColumn)
to just using the top-level functions instead of relying on fields in ActiveDoc, namely docData.
For guessColInfo, this is done by directly passing the small amount of metadata that was previously retrieved from the DocData.
For convertFromColumn, disentangling DocData is a lot more complicated, so instead we construct a fresh DocData object using
the required metadata tables which are now passed in by the data engine.
Test Plan: Existing tests
Reviewers: paulfitz
Reviewed By: paulfitz
Differential Revision: https://phab.getgrist.com/D3913
Summary:
Due to a mishap, two distinct migrations with the same migration
number were introduced into Grist. This diff reconciles them as
best we can, by adding another migration to make sure both desired
changes have run (and running them if not).
Test Plan:
updated a test; checked manually that documents
with different 38 migrations are handled as expected.
Reviewers: georgegevoian, jarek
Reviewed By: georgegevoian, jarek
Differential Revision: https://phab.getgrist.com/D3895
