子宫内膜厚一点好，还是薄一点好？

百度 前天傍晚检测结果出炉，结果显示，传化物流园内加油点加到的柴油，硫含量达到/kg，是国家规定车用柴油国Ⅴ标准（不大于10mg/kg）的40多倍；而废纸回收市场停车场内加油点加到的柴油，硫含量更是达到/kg，是国Ⅴ标准的70倍多!“硫含量严重超标，最直接的后果是汽车排放超标，严重污染空气质量，危害人体健康。

I've just checked in a working postgresql backend for Roundup. That'll make a few people a lot happier.

Along the way, I cleaned up the general rdbms backend implementation. MySQL exhibited some fun behaviour... when running the following:

    set autocommit=0;
    CREATE TABLE ids (id varchar(255)) TYPE=InnoDB;
    CREATE INDEX ids_id_idx on ids(id);
    insert into ids values ('1');
    commit;
    delete from ids where id='1';
    rollback;
    select * from ids where id='1';
    select * from ids where id=1;
    drop table ids;

That will list an empty table on the first select, and the contents of the table on the second select. Note the id column is clearly a varchar. Now the fun part. Commenting out the "CREATE INDEX" statement makes BOTH selects work!

It's a bug in the InnoDB implementation. Switching to BDB makes it work too. BDB is a lot slower though. I've reported it as MySQL bug 1810.

Until it's addressed, I'll be defaulting the table type in the mysql backend to BDB (which is the setting in maint-0-6, thankfully)