This list is still from Linus. MM

The variables should be static.

There should be different error numbers for the different errors instead of
just -1 for them all.
    
Missing library functions to_date et al.
    
Oracle has array operations that enhances speed. When implementing it in
ecpg it is done for compatibility reasons only. For them to improve speed
would require a lot more insight in the postgres internal mechanisms than I
possess.
    
As well as complex types like records and arrays, typedefs would be a good
thing to take care of.
    
To set up a database you need a few scripts with table definitions and other
configuration parameters. If you have these scripts for an old database you
would like to just apply them to get a postgres database that works in the
same way. The functionality could be accomplished with some conversion
scripts. Speed will never be accomplished in this way. To do this you need a
bigger insight in the database construction and the use of the database than
could be realised in a script.

Now comes my list (MM):

The return code is alway -1 in case of an error. You cannot see which error
occured by examining the return code.

ecpg does not understand enum datatypes.

There is no exec sql prepare statement.

The complete structure definition has to be listed inside the declare
section for ecpg to be able to understand it.

There is no way yet to fill a complete array with one call except arrays of
[unsigned] char which are considered strings.

ecpg cannot use pointer variables except [unsigned] char *

give back the number of tuples affected via sqlca

exec sql disconnect {current|default|all|connectionname|connection_hostvar};
 oder <disconnect statement> ::=
    DISCONNECT <disconnect object>

 <disconnect object> ::=
      <connection object>
    | ALL
    | CURRENT
 commit release|commit work release auch disconnect

It is not neccessary to check for "not found" after all commands.