feat: add bft function testcases in new substrait testfile format

[srikrishnak]

No description provided.

[srikrishnak]

The breaking changes check failure is not related to this commit. It pointed to changes in update rel. Once we rebase, it will pass.

[jacques-n]

We should get CI running first for the coverage code before adding these. That way we won't accidentally add broken things.

Second, we should also add the automated tool that does the conversion so other people can use it if they have their own tests.

[EpsilonPrime]

This turned into a review of the BFT source testcases since the conversion process is working correctly.

[EpsilonPrime]

I don't see the count_star test file.

[srikrishnak]

The count_star() function name doesn't match any function in substrait functions. So, unable to add it as coverage breaks.

[EpsilonPrime]

future: we are going to need to revisit all of the string cases once we add collation and character set. (For instance, the capital letter I in Turkish does not become i but a version without a dot.)

[EpsilonPrime]

later: add tests for null and negative counts

[srikrishnak]

We should get CI running first for the coverage code before adding these. That way we won't accidentally add broken things.

Second, we should also add the automated tool that does the conversion so other people can use it if they have their own tests.

Raised a bft PR substrait-io/bft#97 for the script as well.

[srikrishnak]

I took care of review comments but had to rebase because of conflicting file.
After rebase, I new counter in coverage pointed to around 400 signature matches although the testcases would pass on bft.

Below are the different cases which caused signature mismatches even for functions which have the right test cases, addressed almost all of them (4 is not yet fixed). Now I am left with around 250 odd test cases. Now, mostly I see failures which are actual signature mismatches or (4.)

1. tests with return type SubstraitError()
2. tests for functions like or having variadic arguments
3. tests with return type decimal
4. tests with return type any1
5. tests with return type any

[srikrishnak]

I took care of review comments but had to rebase because of conflicting file. After rebase, I new counter in coverage pointed to around 400 signature matches although the testcases would pass on bft.

Below are the different cases which caused signature mismatches even for functions which have the right test cases, addressed almost all of them (4 is not yet fixed). Now I am left with around 250 odd test cases. Now, mostly I see failures which are actual signature mismatches or (4.)

1. tests with return type SubstraitError()
2. tests for functions like or having variadic arguments
3. tests with return type decimal
4. tests with return type any1
5. tests with return type any

raised PR #744 to fix issues with function lookup in coverage tool.
This PR will need rebase on #744

[EpsilonPrime]

There's still the question of extract with microsecond but since that's mirroring the existing test it's fine to investigate it later.

[srikrishnak]

I raised a draft PR substrait-io/bft#98 to check how pipeline runs go with these testcases on bft.

[srikrishnak]

I raised a draft PR substrait-io/bft#98 to check how pipeline runs go with these testcases on bft.

I checked locally by using the DRAFT PR for duckdb, snowflake, postgres and sqllite. Made sure there are no regressions on them.

The previous patch had one offending test case for sqllite, removed that testcase.

[jacques-n]

Great work @srikrishnak and @EpsilonPrime ! Thanks for thorough update and review.