[mdb-dev] Segfault reading JET3 mdb

[Will N. <wi...@gb...>]

Currently I get a segfault when reading a JET3 database I have
here. Unfortunately the file is rather large and proprietary, so I
can't really make it available. It seems that "mdb_crack_row" is
returning a "num_cols" different from the length of "fields" so
subsequent iteration over "fields" runs past the end of the array. The
returned value of "num_cols" is not consistently greater or smaller
than the length of "fields", it seems like the wrong value entirely is
being read. 0.5 didn't do this, but HEAD did.

The attached patch fixes the problem. Why, I'm not too sure.

Exporting my database still has problems:

1) On certain large tables the text in some fields seems to be offset
slightly i.e. the start of one field would like like the end of
another (the last field?). I have these columns:

id,quote_number,declaration_number,new_renewal,agent_code,long_name
17829,"",37124,"41196NEWTUNICKWilliam N","ewbold & George..."

I have abbreviated the data because the table is rather large. The
data should look more like:

17829,37124,41196,"NEW","TUNICK","William Newbold & George..."

That is an approximation as I'm not entirely conversant with the
schema of this legacy data, but it does illustrate the general
appearance of the bug.

2) Memo fields are not being read quite right. WHen exporting a large
table with memo fields (in fact it is the same table as previously
mentioned) I get error messages:

...
offset 731000832 is beyond EOF
offset 419665920 is beyond EOF
offset 733181952 is beyond EOF
offset 1090744320 is beyond EOF
...

And ultimately a segfault. It seems the value of "lval_pg" when
reading non-inline memo fields is incorrect - again it looks like a
pseudo random value rather than any consistent error.


These problems may be related - many of the simple tables in this
database export cleanly. None of the tables with memo fields export
correctly AFAICT.