This is addition to parent revision fixing build failures.

finite is not used anymore in code base. isfinite is part of C99 and we
assume we only support compilers that support C99.

1. Adding LEX::make_item_sysvar() and reusing it
   in sql_yacc.yy and sql_yacc_ora.yy.
   Removing the "opt_component" rule.

2. Renaming rules to better reflect their purpose:
   - keyword to keyword_ident
   - keyword_sp to keyword_label
   - keyword_sp_not_data_type to keyword_sp_var_and_label

   Also renaming:
   - sp_decl_ident_keyword to keyword_sp_decl for naming consistency
   - keyword_alias to keyword_table_alias,
     for consistency with ident_table_alias
   - keyword_sp_data_type to keyword_data_type,
     as it has nothing SP-specific.

3. Moving GLOBAL_SYM, LOCAL_SYM, SESSION_SYM from
   keyword_sp_var_and_label to a separate rule keyword_sysvar_type.
   We don't have system variables with these names anyway.
   Adding ident_sysvar_name and using it in the grammar that needs
   a system variable name instead of ident_or_text.
   This removed a number of shift/reduce conflicts
   between GLOBAL_SYM/LOCAL_SYM/SESSION_SYM as a variable scope and
   as a variable name.

4. Moving keywords BEGIN_SYM, END (in both *.yy fiels)
   and EXCEPTION_SYM (in sql_yacc_ora.yy) into a separate
   rule keyword_sp_block_section, because in Oracle verb keywords
   (COMMIT, DO, HANDLER, OPEN, REPAIR, ROLLBACK, SAVEPOINT, SHUTDOWN, TRUNCATE)
   are good variables names and can appear in e.g. DECLARE,
   while block keywords (BEGIN, END, EXCEPTION) are not good variable names
   and cannot appear in DECLARE.

5. Further splitting keyword_directly_not_assignable in sql_yacc_ora.yy:
   moving keyword_sp_verb_clause out. Renaming the rest of
   keyword_directly_not_assignable to keyword_sp_head,
   which represents keywords that can appear in optional
   clauses in CREATE PROCEDURE/FUNCTION/TRIGGER.

6. Renaming keyword_sp_verb_clause to keyword_verb_clause,
   as now it does not contains anything SP-specific.

   As a result or #4,#5,#6, the rule keyword_directly_not_assignable
   was replaced to three separate rules:
   - keyword_sp_block
   - keyword_sp_head
   - keyword_verb_clause
   Adding the same rules in sql_yacc.yy, for unification.

6. Adding keyword_sp_head and keyword_verb_clause into keyword_sp_decl.
   This fixes MDEV-16244.

7. Reorganizing the rest of keyword related rules into two groups:
  a. Rules defining a list of keywords and consisting of only terminal symbols:
    - keyword_sp_var_not_label
    - keyword_sp_head
    - keyword_sp_verb_clause
    - keyword_sp_block_section
    - keyword_sysvar_type

  b. Rules that combine the above lists into keyword places:
    - keyword_table_alias
    - keyword_ident
    - keyword_label
    - keyword_sysvar_name
    - keyword_sp_decl
  Rules from the group "b" use on the right side only rules
  from the group "a" (with optional terminal symbols added).
  Rules from the group "b" DO NOT mutually use each other any more.
  This makes them easier to read (and see the difference between them).

  Sorting the right sides of the group "b" keyword rules alphabetically,
  for yet better readability.

Fixed by marking NEXTVAL() and LASTVAL() to be replicated row based

Fixing PERIOD to PERIOD_SYM in the %left command.

The failures with valgrind occur as a result of Spider sometimes using the
wrong transaction for operations in background threads that send requests to
the data nodes.  The use of the wrong transaction caused the networking to the
data nodes to use the wrong thread in some cases.  Valgrind eventually
detects this when such a thread is destroyed before it is used to disconnect
from the data node by that wrong transaction when it is freed.

I have fixed the problem by correcting the transaction used in each of these
cases.

Author:
  Jacob Mathew.

Reviewer:
  Kentoku Shiba.

Merged:
  Commit 4d576d9d on branch bb-10.3-MDEV-12900

The failures with valgrind occur as a result of Spider sometimes using the
wrong transaction for operations in background threads that send requests to
the data nodes.  The use of the wrong transaction caused the networking to the
data nodes to use the wrong thread in some cases.  Valgrind eventually
detects this when such a thread is destroyed before it is used to disconnect
from the data node by that wrong transaction when it is freed.

I have fixed the problem by correcting the transaction used in each of these
cases.

Author:
  Jacob Mathew.

Reviewer:
  Kentoku Shiba.

Merged:
  Commit 4d576d9d on branch bb-10.3-MDEV-12900

The failures with valgrind occur as a result of Spider sometimes using the
wrong transaction for operations in background threads that send requests to
the data nodes.  The use of the wrong transaction caused the networking to the
data nodes to use the wrong thread in some cases.  Valgrind eventually
detects this when such a thread is destroyed before it is used to disconnect
from the data node by that wrong transaction when it is freed.

I have fixed the problem by correcting the transaction used in each of these
cases.

Author:
  Jacob Mathew.

Reviewer:
  Kentoku Shiba.

- Using array_elements() instead of a constant to iterate through an array
- Adding some comments
- Adding new-line function comments
- Using STRING_WITH_LEN instead of C_STRING_WITH_LEN

- Removing the unused "THD*" paramer from Lex_ident_sys_st::to_size_number()
- Removing redundant #include for "sql_tvc.h" and "vers_utils.h"

Symlink core from default location (/cores/core.%P), so that lldb can
handle it.

don't rely on imprecise my_file_opened | my_stream_opened,
scan the array for open handles instead.

also, remove my_print_open_files() and embed it in my_end()
to have one array scan instead of two.

Change the following to statistic counters:
* my_file_opened
* my_file_total_opened
* my_stream_opened
* my_tmp_file_created

There is one non-statistics use of my_file_opened/my_stream_opened
in my_end which prints a warning if we shutdown and its still open.
It seems excessive to hold locks to prevent this warning.

A file descriptor is already a unique element per process - in Windows,
protection occurs at fd allocation using THR_LOCK_open in my_win_{,f}open
and in other OSes, a unique fd to file map exists at the OS level.
So accesses to my_file_info[fd] don't need to be protected by the
THR_LOCK_open.

my_close/my_fclose where restructured to clear out the my_file_info
before the close/my_win_close/my_win_fclose. After these calls another
thread could gain the same file descriptor. So for Windows this
the file_info elements available to the my_win_{,f}_open are released
during the invalidate_fd call within my_win_close. No locking is needed
as the my_win_{,f}open is searching for a invalidate entry which is
determined by a single value change.

my_fclose also changed for non-Windows to retry closing if EINTR was
returned, same as my_close.

Closes #657

Closes #639

it's redundant. remove it, use my_error() directly.

simplify. move common code inside, specify common flags inside,
rewrite dead code (`if (mode & O_TEMPORARY)` on Linux, where
`O_TEMPORARY` is always 0) to actually do something.

In the spirit of the man page "Never use this function."
lets purge off this implementation. mkstemp is a widely
available alternative.

Closes #661.

Print error message and return when fopen fails.

Closes #634

for consistency with EVP_CIPHER_CTX_new() and EVP_CIPHER_CTX_free().

As the EVP_DIGESTINIT(3) man page says:

EVP_MD_CTX_create() and EVP_MD_CTX_destroy() were renamed to
EVP_MD_CTX_new() and EVP_MD_CTX_free() in OpenSSL 1.1.

Closes #621

Symlink core from default location (/cores/core.%P), so that lldb can
handle it.

don't rely on imprecise my_file_opened | my_stream_opened,
scan the array for open handles instead.

also, remove my_print_open_files() and embed it in my_end()
to have one array scan instead of two.

Change the following to statistic counters:
* my_file_opened
* my_file_total_opened
* my_stream_opened
* my_tmp_file_created

There is one non-statistics use of my_file_opened/my_stream_opened
in my_end which prints a warning if we shutdown and its still open.
It seems excessive to hold locks to prevent this warning.

A file descriptor is already a unique element per process - in Windows,
protection occurs at fd allocation using THR_LOCK_open in my_win_{,f}open
and in other OSes, a unique fd to file map exists at the OS level.
So accesses to my_file_info[fd] don't need to be protected by the
THR_LOCK_open.

my_close/my_fclose where restructured to clear out the my_file_info
before the close/my_win_close/my_win_fclose. After these calls another
thread could gain the same file descriptor. So for Windows this
the file_info elements available to the my_win_{,f}_open are released
during the invalidate_fd call within my_win_close. No locking is needed
as the my_win_{,f}open is searching for a invalidate entry which is
determined by a single value change.

my_fclose also changed for non-Windows to retry closing if EINTR was
returned, same as my_close.

Closes #657

Closes #639

it's redundant. remove it, use my_error() directly.

simplify. move common code inside, specify common flags inside,
rewrite dead code (`if (mode & O_TEMPORARY)` on Linux, where
`O_TEMPORARY` is always 0) to actually do something.

In the spirit of the man page "Never use this function."
lets purge off this implementation. mkstemp is a widely
available alternative.

Closes #661.

Print error message and return when fopen fails.

Closes #634

for consistency with EVP_CIPHER_CTX_new() and EVP_CIPHER_CTX_free().

As the EVP_DIGESTINIT(3) man page says:

EVP_MD_CTX_create() and EVP_MD_CTX_destroy() were renamed to
EVP_MD_CTX_new() and EVP_MD_CTX_free() in OpenSSL 1.1.

Closes #621