SQL statement terminator should be semicolon.

[kyukhin]

In order to have multi-line SQL statements, and avoid confusion let's make ;mandatory statement terminator.

I am able to say
select * from ww
there is no semicolon. This might look
convenient, but what is somebody implements
a dumb client?
Currently, If I have a multi-line statement,
there's no error
"
tarantool> ocelot_conn:execute('select * from
> ww')
--
...
"
But that won't last -- people will want to
type in SQL, e.g.
tarantool-sql> select * from
> ww
If ";" isn't the end-of-statement delimiter,
and newline is a signal for execution, then
the dumb client will try to execute
"select * from" and fail.
Of course, it's always hard to know when an
SQL statement is really supposed to end -- only
a sophisticated client can detect that.
But probably it would help if statements
had to end with semicolons.

[kshcherbatov]

This functionality seems to be already in Tarantool:
There is "pragma delimiter" introduced woth @GeorgyKirichenko patch c3afd0 "Add support for backslash commands in console".
Usage scenario:

tarantool> \set delimiter ;
---
...

tarantool> \set language sql
---
- true
...

tarantool> CREATE TABLE t2  (a int primary key,
         > b int)
         > ;

[Gerold103]

tarantool> \set delimiter ;
---
...

This must be default behaviour. A user has not to set delimiter ; manually.
And please, test, what will happen, for example, on a request like this:

select 'select "abcd";';

[Gerold103]

I have found two problems:

1. When I use arrows on my keyboard to get a previous statement, its delimiter ; is truncated:

tarantool> select 'select "abcd";';
---
- - ['select "abcd";']
...

tarantool> select 'select "abcd";'

Here I got select 'select "abcd";' by pressing keyboard button 'arrow up'. I think, that such history is unusable. It must contain terminating ; as well.

2. Spaces after ; surpass it.

tarantool> select 'select "abcd";';     
         > 

Here I put several spaces after ; and disabled its termination meaning.

[kshcherbatov]

Thank you for the comments,
as for first comment I've accounted it in the patch bellow.

diff --git a/src/box/lua/console.lua b/src/box/lua/console.lua
index e7cb50a..7644d33 100644
--- a/src/box/lua/console.lua
+++ b/src/box/lua/console.lua
@@ -67,6 +67,7 @@ local function set_language(storage, value)
         local msg = 'Invalid language "%s", supported languages: lua and sql.'
         return error(msg:format(value))
     end
+    set_delimiter(storage, value == 'sql' and ';' or nil)
     storage.language = value
     return true
 end
@@ -316,7 +317,11 @@ local function local_read(self)
                 break
             end
         elseif #buf >= #delim and buf:sub(#buf - #delim + 1) == delim then
-            buf = buf:sub(0, #buf - #delim)
+            local sub_len = #buf - #delim
+            if box.session.language == 'sql' and delim == ';' then
+                sub_len = sub_len + 1
+            end
+            buf = buf:sub(0, sub_len)
             break
         end
         buf = buf.."\n"
@@ -354,8 +359,12 @@ local function client_read(self)
         -- Escape sequence to close current connection (like SSH)
         return nil
     end
-    -- remove trailing delimiter
-    return buf:sub(1, -#delim-1)
+    -- remove trailing delimiter if it is not SQL ;
+    if box.session.language == 'sql' and delim == ';' then
+        return buf
+    else
+        return buf:sub(1, -#delim-1)
+    end
 end

As for second, we should trim every new data portion on read from console or socket using smth like this:

function rtrim(s)
  local n = #s
  while n > 0 and s:find("^%s", n) do n = n - 1 end
  return s:sub(1, n)
end

But I am not shure that it is a good idea.

[Gerold103]

Please, use the mailing list to send patches. I can not review them here.

[kshcherbatov]

won't implement for now
@kostja confirmed