[tarantool-patches] Re: [PATCH] sql: ban of REINDEX syntax

[Alexander Turenko]

Ivan, please consider decisions below, update the commit and send it
again.

WBR, Alexander Turenko.

On Wed, Mar 21, 2018 at 07:42:52PM +0300, Kirill Yukhin wrote:
> On 21 мар 19:08, Alexander Turenko wrote:
> > Kirill, can you suggest best ways to go forward on detable questions
> > below?
> I think there're no specific ways to raise such. If you'd like to bring
> my attention: CC me explicitly in e-mail.
> 
> > 
> > On Wed, Mar 21, 2018 at 06:37:38PM +0300, n.pettik wrote:
> > > Hello. See comments below.
> > > 
> > > >This change cut
> > > 
> > > Misspelling: cut(s).
> > > 
> > > >The reason described in 'parse.y' in comment related
> > > >to '%fallback'.
> > > 
> > > Don't reffer to comments in your commit message. Instead, provide
> > > short desctiption/explanation.
> > > 
> > 
> > Why we need to duplicate explanation? A commit message is a part of the
> > commit.
> You might break commit-comment connection by commits that follows. 
> 

Proposed to describe it shortly in the commit message like so:

Note that REINDEX keyword is still in '%fallback' directive in
'parse.y', because we need at least one usage of a keyword in 'parse.y'
to prevent exclusion from 'parse.h' (it would break the build).

> > > >+++ b/src/box/sql/parse.y
> > > >@@ -202,6 +202,17 @@ columnname(A) ::= nm(A) typetoken(Y). {sqlite3AddColumn(pParse,&A,&Y);}
> > > >// fallback to ID if they will not parse as their original value.
> > > >// This obviates the need for the "id" nonterminal.
> > > >//
> > > >+// A keyword is checked for being a reserve one in `nm`, before
> > > >+// processing of this %fallback directive. Reserved keywords included
> > > >+// here to avoid the situation when a keyword has no usages within
> > > >+// `parse.y` file (a keyword can have more or less usages depending on
> > > >+// compiler defines). When a keyword has no usages it is excluded
> > > >+// from autogenerated file `parse.h` that lead to compile-time error.
> > > >+//
> > > >+// See [1] for more information.
> > > >+//
> > > >+// [1]: https://www.sqlite.org/src/info/007aec11333432e0
> > > >+//
> > > 
> > > I would better put link to commit message or to comments of original
> > > issue. I stick to the point that source code doesn't seem to be good
> > > place for links. Also, in Tarantool we use C-like commenting style.
> > > 
> > 
> > I don't see nothing bad with using links, epsecially when it can
> > simplify reader experience.
> IMHO, internet is volatile. WWW-links are not verbatim enough to me, and
> hence are fallcious artifacts to put into comment. I'd defer final decision
> to 1st level reviewer.
> 

Link with SHA1 is better IMO than just SHA1, but I understood that the
point is about our feelings of the code we write. Proposed to include it
like so:

See fossil check-in [007aec11] or CVS commit 2045 in SQLite for more
information.