Re: IF (NOT) EXISTS in psql-completion

Hello, this is the new version.

At Tue, 13 Sep 2016 10:50:13 +0900 (Tokyo Standard Time), Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp> wrote in <20160913.105013.65452566.horiguchi.kyotaro@lab.ntt.co.jp>
> > > This patch consists of the following files. Since these files are
> > > splitted in strange criteria and order for historical reasons,
> > > I'll reorganize this and post them later.

The focus of this patch has changed. The first motivation was
completing IF-EXISTS but the underlying issue was flexibility of
psql_completion. And as Pavel's suggestion, keywords suggested
along with database objects should follow the character case of
input.

For the purpose of resolving the issues, I reorganized the
confused patch set. The attached patches are organized as the
following.

1. Refactoring tab-complete to make psql_completion code

  Does two things. One is moving out the macros that has grown to
  be too large to stay in tab_completion.c to new file
  tab-complete-macros.h The other is separating out the else-if
  sequence in psql_completion() as a new function
  psql_completion_internal(). This allows us to the following
  things.

  - Exit from arbitrary place in the former-else-if sequence just
    by return.

  - Do other than "if(matching) { completion }" in anywhere
    convenient in the midst of the former-els...

  - Recursively matching for sub syntaxes. EXPLAIN, RULE and
    others are using this feature. (Needs the 4th patch to do
    this, though)

2. Make keywords' case follow to input

  Allow the keywords suggested along with databse objects to
  follow the input letter case. The core part of this patch is a
  new function additional_kw_query(), which dynamically generates
  additional query string with specified keywords in the desired
  letter case. COMPLETE_WITH_* macros are modified to accept the
  function.

3. Fix suggested keywords to follow input in tab-completion session 2

  The 2nd patch above leaves some query string containing static
  keyword strings, which results in failure to follow input
  letter cases. Most of them are naturally removed but role names
  are a bother. This patch puts additional query strings for
  several usage of roles but it might be overdone.

4. Introduce word shift and removal feature to psql-completion

  This is the second core for the flexibility of completion code.
  The word shift feature is the ability to omit first several
  words in *MatchesN macros. For example this allows complete
  create-schema's schema elements in a natural code. (Currently
  those syntaxes that can be a schema elements are using
  TailMatches instead of Matches, as the result HeadMatches are
  not available there). The words removing feature is the ability
  to (desructively) clip multiple suceessive words in the
  previous_words list. This feature allows suceeding completion
  code not to care about the removed words, such like UNIQUE,
  CONCURRENTLY, VERBOSE and so on.

5. Add suggestion for IF (NOT) EXISTS for some syntaxes

  This adds IF (NOT) EXISTS suggestion, as a PoC.  This patch no
  loger covers all adoptable syntaces since the places where more
  than boilerplating is required are omitted.

regards,

--
Kyotaro Horiguchi
NTT Open Source Software Center