parse TIMESTAMP{ADD,DIFF} functions

[slanning]

Otherwise, get something like this error:
vtgate.go:1091] Execute: symbol MINUTE not found in table or subquery, request:
map[Sql:SELECT TIMESTAMPDIFF(MINUTE, ....

reference:
https://dev.mysql.com/doc/refman/5.7/en/date-and-time-functions.html#function_timestampadd

Signed-off-by: Scott Lanning scott.lanning@booking.com

[slanning]

One failed, not sure if it's something I should fix.

[deepthi]

One failed, not sure if it's something I should fix.

Yes, it should be fixed. Not sure why goyacc is not producing the newline locally vs in travis, though.

[sougou]

Change looks good. One minor nit.

[sougou]

New tokens need to be added to either reserved_keywords or non_reserved_keywords. In this case, I think it makes sense to add them to non_reserved_keywords.

[slanning]

Oh right, not sure how I forgot that.

[sougou]

Still waiting on this. I can merge this once you've added them.

[slanning]

Sorry, I was OoO. I had to put it in reserved_keywords to avoid reduce/shift conflicts:

289: shift/reduce conflict (shift 54(0), red'n 822(0)) on '('
state 289
        function_call_nonkeyword:  TIMESTAMPADD.openb sql_id ',' value_expression ',' value_expression closeb 
        non_reserved_keyword:  TIMESTAMPADD.    (822)

        '('  shift 54
        .  reduce 822 (src line 3359)

        openb  goto 564

290: shift/reduce conflict (shift 54(0), red'n 823(0)) on '('
state 290
        function_call_nonkeyword:  TIMESTAMPDIFF.openb sql_id ',' value_expression ',' value_expression closeb 
        non_reserved_keyword:  TIMESTAMPDIFF.    (823)

        '('  shift 54
        .  reduce 823 (src line 3360)

        openb  goto 565

I don't pretend to understand it, but for example CURRENT_TIMESTAMP is also in (function_call_nonkeyword and) reserved_keyword.

[slanning]

(edit: replying to @deepthi - not sure what's going on in the github interface)

Yes, it should be fixed. Not sure why goyacc is not producing the newline locally vs in travis, though.

I occasionally wonder why sql.go is even included in the repo (I usually wonder this when merge conflicts happen, which is usually due to someone else also changing sql.y).

I noticed that when I ran goyacc without my changes, the only change was removing that one empty line. Not sure if it's a version difference or what.

[sougou]

I occasionally wonder why sql.go is even included in the repo (I usually wonder this when merge conflicts happen, which is usually due to someone else also changing sql.y).

Go has a guideline that requires all generated files must be checked in. They want to maintain that an end user should be able to build the project without using make.

I noticed that when I ran goyacc without my changes, the only change was removing that one empty line. Not sure if it's a version difference or what.

This is probably a version difference, but in gofmt. Can you update to go1.11 and retry?

[slanning]

That's what I use actually:

go version go1.11.4 linux/amd64

So, dunno.

[slanning]

Turns out my goyacc was old (from September). I think I hadn't rerun bootstrap.sh since upgrading the go version..

[slanning]

I --force pushed to the branch without the commit, and it closed this PR, then I reopened it after pushing with the commit. Not sure if the various checks will be happy.

[sougou]

Overriding failed DCO.