btr_pcur_store_position() and btr_pcur_restore_position() will not
save the extra header associated with "instanly added" columns,
because the stored and restored prefixes only ever include key
columns, which cannot be added instantly.

For ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPACT records, we have to
identify whether the record contains more columns than when the
clustered index tree was originally created. In this case, the
number of fields
(FIXME: number_of_added_fields-1, not the total number of fields)
needs to be written to the record header.

rec_comp_status_t: An enum of the status bit values.

rec_leaf_format: An enum that replaces the bool parameter of
rec_init_offsets_comp_ordinary().

rec_get_converted_size_comp(): Do not support infimum,supremum.
They are never supposed to be stored in dtuple_t, as page creation
nowadays uses a lower-level method for inserting them.

innobase_add_instant_try(): Initialize the dtuple_t::info_bits
appropriately to REC_INFO_MIN_REC_FLAG | REC_STATUS_COLUMNS_ADDED
so that it can be inserted into a ROW_FORMAT!=REDUNDANT table.

row_merge_read_clustered_index(), IndexPurge::open():
Before testing for the 'default row' pseudo-record, advance
the cursor from the infimum pseudo-record, and then move back
if needed.

rec_offs_any_flag(): The EXTERNAL and DEFAULT flags are only mutually
exclusive for individual fields. A record can contain both EXTERNAL and
DEFAULT fields, as long as they are separate fields.

sel_restore_position_for_mysql(): Pass the correct index to
rec_is_default_row(). (We should really remove btr_cur_t::index
and use page_cur_t::index isntead.)

This reverts commit d198f961.

The 'default row' record (which stores the default values of
instantly added columns when they were added) must be ignored
during normal operation.

Because btr_pcur_store_position() can theoretically be invoked
on the pseudo-record, we must check every caller of
btr_pcur_restore_position(). Similarly, when positioning the
cursor to the start of an index with btr_pcur_open_at_index_side(),
or when moving to the previous record, we must be careful to
skip the 'default row' record.

rec_is_default_row(): Check whether a leaf page record is the
'default row' pseudo-record.

row_import_set_sys_max_row_id(): Check for a table that is empty but
contains the 'default row' record.

IndexPurge::open(), row_merge_read_clustered_index(),
dict_stats_analyze_index_level(), row_sel_try_search_shortcut(),
row_sel(), sel_restore_position_for_mysql(), row_search_mvcc():
Skip the 'default row' pseudo-record.

fts_get_max_doc_id(): Assert that the 'default row' cannot exist
in the secondary index FTS_DOC_ID_INDEX(FTS_DOC_ID).

page_find_rec_max_not_deleted(): Assert that we are operating
on a leaf page. Also skip the 'default row' pseudo-record, not
only delete-marked records.

row_search_get_max_rec(): Assert that page_find_rec_max_not_deleted()
did not return any delete-marked or 'default row' record.

With instant ADD COLUMN, rec_get_offsets() must reliably know if it
is dealing with a user record (not the page infimum/supremum).

The 'default row' record (which is the only search tuple and the only
leaf-page record that carries the REC_INFO_MIN_REC_FLAG)  must be searched
by using the conventional means. Do not slow down the search shortcut by
taking the MIN_REC_FLAG into account.

rec_offs_validate(): Do not dereference index==NULL.

rec_validate(): Properly handle missing columns.

dict_table_get_nth_v_col(): Assert that no default value is ever set
on a virtual column. Virtual columns are not handled by the
instant ADD COLUMN mechanism.

ha_innobase_inplace_ctx::commit_instant(): Copy the default values
from instant_table to old_table.
Take deep copy of the dict_table_t::v_cols[].

instant_add_instant_try(): Copy the default values to instant_table.

dict_index_get_nth_field_def(): Define the function in the .h file.

dict_col_t::def_val.data: Change the data type to const void*
to reduce the number of type casts.

This could have been done as part of MDEV-11487, which removed the
support for InnoDB internal temporary tables.

The column SYS_INDEXES.MERGE_THRESHOLD was added in MySQL 5.7.
This makes the system table the very first one where
instant ADD COLUMN has been used.

rec_get_nth_cfield(): Remove the heap parameter.
Copying the default value of a field into a heap is only
needed in row_build_low(type=ROW_COPY_DATA).

rec_get_nth_field(): Use the original definition.
There are very few rec_get_nth_cfield() calls compared to
rec_get_nth_field().

Assert that the flags EXTERNAL, DEFAULT, and NULL are mutually exclusive.

Store dict_col_t::def_val inline, using an anonymous struct.

dict_mem_table_fake_nth_col_default(),
dict_col_set_added_column_default(): Remove.

dict_mem_fill_column_struct(): Initialize def_val.

rec_get_offsets(): Use the 'leaf' parameter for non-debug purposes.
Initialize all fields for ROW_FORMAT=REDUNDANT records that precede
an instant ADD COLUMN.

rec_offs_make_valid(): Add the parameter 'leaf'.

rec_copy_prefix_to_dtuple(): Assert that the tuple is only built
on the core fields. Instant ADD COLUMN only applies to the
clustered index, and we should never build a search key that has
more than the PRIMARY KEY and possibly DB_TRX_ID,DB_ROLL_PTR.
All these columns are always present.

dict_index_build_data_tuple(): Remove assertions that would be
duplicated in rec_copy_prefix_to_dtuple().

rec_get_n_nullable(): Get the number of nullable columns.

rec_init_offsets(): Support ROW_FORMAT=REDUNDANT records that
were written before instant ADD COLUMN.

cmp_rec_rec_with_match(): Implement comparison of MIN_REC_FLAG records.

Create each table with a SPATIAL INDEX, because when MySQL 5.7
(and MariaDB 10.2.2) introduced SPATIAL INDEX for InnoDB tables,
a table-rebuilding LOCK=NONE would be refused if the table
contains any SPATIAL INDEX. With instant ADD COLUMN, the operations
must be allowed, and must report '0 rows affected' for everything
else except ROW_FORMAT=COMPRESSED tables.

FIXME: When instant ADD COLUMN is enabled, the ALGORITHM=INPLACE
will have to be changed to LOCK=NONE.

followup for b5ead3a6

libmariadb3 does not replace 10.1's libmariadbclient18 anymore,
but it still conflicts with it, both install dialog.so

This fixes a regression that only affects debug builds, caused by
commit 48192f96 which is necessary
preparation for MDEV-11369 instant ADD COLUMN. (Although that is a
10.3 task, to ease merges between 10.2 and 10.3, this change that
improves debug checks was pushed to 10.2 already.)

Unlike btr_pcur_restore_position(), rtr_cur_restore_position()
can create a search tuple out of a non-leaf page record. So,
we must pass 'bool leaf' parameter to dict_index_build_data_tuple().

dict_index_build_data_tuple(): Add a debug-only parameter 'bool leaf'.

rec_copy_prefix_to_dtuple(): Make the parameter debug-only.

row_sel_get_clust_rec_for_mysql(): In the debug code for spatial index,
remove an unnecessary call to buf_page_get_gen(), and use the already
latched block directly.

MDEV-13851 Always check table options in ALTER TABLE…ALGORITHM=INPLACE

In the merge of MySQL 5.7.9 to MariaDB 10.2.2, some code was included
that prevents ADD SPATIAL INDEX from being executed with ALGORITHM=INPLACE.

Also, the constant ADD_SPATIAL_INDEX was introduced as an alias
to ADD_INDEX. We will remove that alias now, and properly implement
the same ADD SPATIAL INDEX restrictions as MySQL 5.7 does:

1. table-rebuilding operations are not allowed if SPATIAL INDEX survive it
2. ALTER TABLE…ADD SPATIAL INDEX…LOCK=NONE is not allowed

ha_innobase::prepare_inplace_alter_table(): If the ALTER TABLE
requires actions within InnoDB, enforce the table options (MDEV-13851).
In this way, we will keep denying ADD SPATIAL INDEX for tables
that use encryption (MDEV-11974), even if ALGORITHM=INPLACE is used.

This should affect debug builds only. Debug builds will check that
the status bits of ROW_FORMAT!=REDUNDANT records match the is_leaf
parameter.

The only observable change to non-debug should be the addition of
the is_leaf parameter to the function rec_copy_prefix_to_dtuple(),
and the removal of some calls to update the adaptive hash index
(it is only built for the leaf pages).

This change should have been made in MySQL 5.0.3, instead of
introducing the status flags in the ROW_FORMAT=COMPACT record header.

* remove dead code

* rename method to reduce confusion