Issues observed during prettifying about pglast HOT 13 CLOSED

I have found several examples that trigger the "safety belt"

Thanks for the report: I will try to address all of them, as soon as I'll restart working on v2, hopefully next week.

from pglast.

@lelit I'm not sure I understand, we already have them here https://github.com/lelit/pglast/blob/v2/pglast/keywords.py don't we ?

from pglast.

I'm not familiar enough with the parser to know what it implies for a keyword to be in one particular set or another. It would seem like depending on the context, you should check some sets and not others...

from pglast.

@tscholak, I think I cured both problems with the recent commits, any chance you can try current master and (possibly) confirm?

Note to myself: I should spend a tad of time learning the new "GH Actions"...

from pglast.

Thanks, I missed that, sorry. Can you try again with latest master please?

from pglast.

Great, released as v1.12.

from pglast.

@rdunklau, the right vs "right" issue appears to require the list of reserved words: maybe there's a way in libpgquery to expose the list defined in src/include/parser/kwlist.h? I could not spend enough time figuring out how they are actually stored, see keywords.c...

from pglast.

Yes, sorry, one should never read (his own!) source code early in the morning, before first coffee!

from pglast.

That problem should be fixable by considering also TYPE_FUNC_NAME_KEYWORDS, instead of only RESERVED_KEYWORDS here. We should add a test to verify the behaviour with all different kinds of keyword (that is, checking also some COL_NAME_KEYWORDS and UNRESERVED_KEYWORDS...)

from pglast.

Thank you very much for the quick response!
I was not aware of the v2 effort. Are you saying that these issues are already fixed in the v2 branch?

Btw, the queries I am considering are from an open source dataset called Spider, https://yale-lily.github.io/spider. There are almost 10k queries in this dataset, and I think it could be a great asset for your work if you wanted to run a lot of there-and-back-again tests.

from pglast.

I was not aware of the v2 effort. Are you saying that these issues are already fixed in the v2 branch?

No, v1 and v2 are very close in functionality, but the latter targets PG12.

Btw, the queries I am considering are from an open source dataset called Spider, https://yale-lily.github.io/spider. There are almost 10k queries in this dataset, and I think it could be a great asset for your work if you wanted to run a lot of there-and-back-again tests.

Interesting indeed, thanks.

from pglast.

Hi @lelit, thank you very much for looking into this. The two changes you made fixed the examples reported above, well done! I'm very happy :)

What remained an issue though are nested intersect, union, and except blocks as mentioned in the last paragraph of my initial report.

Can you have a look please at situations like these:

(
    SELECT * FROM "table"
) INTERSECT (
    (
        SELECT * FROM "table"
    ) UNION (
        SELECT * FROM "table"
    )
)

(
    (
        SELECT * FROM "table"
    ) UNION (
        SELECT * FROM "table"
    )
) INTERSECT (
    SELECT * FROM "table"
)

etc.?

from pglast.

tried it, works as expected!!!

Thanks so much! Consider this issue closed.

from pglast.