[mdb-dev] [patch] Re: Proposed changes to HACKING doc

[Nirgal V. <con...@ni...>]

I tried to make a patch out of all that, since there is some really good stuff, and I believe it's easier for Brian to merge.

I discarded a few things, like the renaming of fields, because it would not match the source code anymore. It's complex enough reading it... HACKING does help.

On Monday 28 February 2011 08:25:17 dw wrote:
> 8) The definitions for index information seem to be wrong.
> (...)
> Lastly
> | ???? | 1 byte | primary_key | 0x01 if this index is primary |
> Should be
> | ???? | 1 byte | index_flags | see MDB_IDX_* flags |

I disagree with that one.
The flags are above, in the jet3 39 bytes index blocks, ot jet 4 52 bytes one.

> So, when walking the list of index definitions, you use num_idx to tell 
> you how many indices there are.  Then for each index, you use 
> index_Cols_index to index into the column definitions table to find the 
> columns for that index.  The reason num_idx and num_idx_col might be 
> different is not because some of the indices are "real" (whatever that 
> was supposed to mean), but simply because more than one index uses the 
> same column list.  I've seen this happen if there is one index that 
> includes nulls, and one that doesn't.

FYI
Type 2 extra "indexes" are the reverse of contraints, probably used to optimized cascade updates.
These ones don't have matching columns definitions blocks.

> I have a re-written mdb_read_indices that (I believe) correctly parses 
> this data if anyone is interested.

It wouldn't hurt to publish that to the list, would it?
I did that several times, and often had bad surprises, like when there are so many indexes that index definitions cover several DB pages...