Add a helper function for connecting to Snowflake.

[atheriel]

This commit introduces a new odbc::snowflake() function intended to help setting up connections to Snowflake and handling auth correctly on platforms that provide Snowflake-native OAuth credentials. It generally follows the pattern established with odbc::databricks().

We've been testing this internally on Snowflake's container platform, and it works as expected.

[atheriel]

Tests are currently failing because we now have both a Snowflake.Rd and a snowflake.Rd file. I'm not sure what you'd suggest in this case, but happy to fix what's needed.

[simonpcouch]

Ah, interesting. From my perspective, I'd say we transition existing @rdname Snowflakes to @rdname driver-Snowflake and allow ?snowflake to head to docs on the wrapper implemented in this PR.

[atheriel]

@simonpcouch Done.

[hadley]

Since we just added snowflake to our CI, should we add a couple of very basic tests?

[atheriel]

I've added a number of unit tests (and fixed some issues in the process!), but due to the design of the current CI secrets -- which supply entire connection strings -- I can't see an easy way to pass real credentials through the new odbc::snowflake() function.

[hadley]

@simonpcouch could you please take a look at how we're managing secrets to see if we could facilitate their use here?

[hadley]

I'd be tempted to document them alltogether, adding a new section that details the snowflake method variations. But might be easiest for me to do this after the PR is merged.

[simonpcouch]

Just pushed a quick demo of using the secret in tests--fine by me to have both the PWD embedded in a connection string and the PWD itself in the GITHUB_ENV. The remaining connection information is in .github/odbc/odbc.ini.

[chendaniely]

quarto render seems to be unable to connect to snowflake. Manually running the code chunk and runnign the code in the console work just fine.
but pressing the render button seems to break:

processing file: reprex.qmd
  |...................................                 |  67% [unnamed-chunk-1]
Quitting from lines  at lines 7-20 [unnamed-chunk-1] (reprex.qmd)
Error:
! nanodbc/nanodbc.cpp:1139: 00000
[RStudio][DSI] (20032) Required setting 'UID' is not present in the connection settings. 
Backtrace:
  1. DBI::dbConnect(...)
  2. DBI::dbConnect(...)
  3. odbc (local) .local(drv, ...)
  6. odbc::dbConnect(...)
  7. odbc (local) .local(drv, ...)
  8. odbc:::OdbcConnection(...)
 10. odbc:::odbc_connect(...)
                                                                                                            
Execution halted

reprex quarto file using the cybersyn webtraffic database: https://docs.cybersyn.com/consumer-insights/web-traffic-foundation

---
title: "Untitled"
format: html
---

```{r}
library(dplyr)
library(tidyr)
library(DBI)
library(dbplyr)
library(odbc)

conn <-
  DBI::dbConnect(
    odbc::snowflake(),
    warehouse = "DEVREL_WH_LARGE",
    database = "WEB_TRAFFIC_FOUNDATION_EXPERIMENTAL",
    schema = "CYBERSYN"
  )
```

[simonpcouch]

In the above, "ambient" credentials are those set by default in the Posit Workbench build hosted at dev palm ptd:

grep("SNOWFLAKE", names(Sys.getenv()), value = TRUE)
#> [1] "SNOWFLAKE_ACCOUNT" "SNOWFLAKE_HOME"    "SNOWFLAKE_ROLE"  

If this is just a config issue, we should detect the conditions when this error will be raised and raise a more informative one before dispatching to dbConnect.OdbcConnection().

EDIT: noting that I can reproduce Daniel's issue.

[atheriel]

This is ready for re-review:

• Added additional tests, including a smoke test to make sure that odbc::snowflake() can be used to connect to the real Snowflake account in our CI system.
• I've fixed the Quarto rendering issue by removing the call to in-process RStudio APIs; instead we parse the Workbench-generated file directly, which should be a lot more portable and future-proof.
• CI failures seem to be related to code I haven't changed, but I may be mistaken.

[simonpcouch]

Looking solid! That Snowflake CI failure for the "real" connection did indeed crop up elsewhere and is uninformative due to #798—no need to address here. I do see, in R CMD check runs:

❯ checking Rd \usage sections ... WARNING
  Undocumented arguments in Rd file 'driver-Snowflake.Rd'
    ‘catalog_name’ ‘schema_name’

Could you look into those?

Submitting review as "Comment" rather than "Approve" only so as to defer to Hadley on the timeline to merge this one. :)

[atheriel]

The warning about missing docs is unrelated to this PR but I fixed it anyway to try and make CI happy :)

[garrettgman]

cc @skaltman

[simonpcouch]

The GHA failure:

── Failed tests ────────────────────────────────────────────────────────────────
Error ('test-driver-snowflake.R:2:3'): can connect to snowflake
Error in `test_con("SNOWFLAKE")`: ! ODBC failed with error 00000 from [Snowflake][Snowflake].
*  Password not found.
i From 'nanodbc/nanodbc.cpp:1147'.

...is due to the relevant repository secret not being available and is fine to ignore. Game to merge with your thumbs-up, @atheriel—just let us know whether you want to address the warning -> error conversion review comment in this PR or file a separate issue to make that happen.